refactor: 移除 context-mode，Chrome 快照改为直接返回

context-mode 的 ctx_execute 系工具以真实 shell 执行且权限为 allow，绕过了
bash 策略与 protectedWrites；其每轮注入、独立 MCP 子进程与 skills 也与组合包
提示词单一所有权冲突。大输出职责已由 pi-overflow（入口截断）与
pi-compaction（压力遮蔽）承担：
- 删除 context-mode 依赖、wrapper/过滤器、skills，以及 ctx_* 的权限、
  tools.jsonc、路由规则、tool-search 常驻与 4 个分组
- better-sqlite3@12 仅为 context-mode 传递依赖，随之移除根 allowScripts
- pi-chrome：按用户决定移除快照隐私规则与 .pi/chrome-context 临时文件机制，
  快照 JSON 直接作为工具结果返回、details 只留小型元数据，超长由 pi-overflow
  截断并用 overflow_query 提取；授权生命周期不变

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

AGENTS.md

@@ -18,10 +18,10 @@
 | `pi-hermes-memory/` | Hermes 风格持久记忆：policy-only 注入、六记忆工具、`skill_manage` |
 | `pi-minimal-footer/` | 紧凑 footer：上下文仪表、订阅用量条、扩展状态 |
 | `pi-notify/` | Kitty OSC 99 完成与权限确认通知，精确聚焦来源窗口 |
-| `pi-compaction/` | 增量 Session segment、确定性压缩、后台语义摘要、effective token、native compaction |
+| `pi-compaction/` | 压力触发的观察遮蔽（Pi context_edit）、阈值整段摘要、effective token、native compaction |
+| `pi-overflow/` | 新工具结果统一截断：超阈值原文落盘 + 有界视图 + `overflow_query` |
 | `extensions/fff-override.ts` | FFF `override` 模式接管 `find`/`grep`/`multi_grep` 与 `@` 补全 |
 | `extensions/codegraph.ts` | 原生 `codegraph_explore`，自有 stdio JSON-RPC 直连 `codegraph serve`（`codegraph-client.ts`） |
-| `extensions/context-mode.ts` | 经 `prompt-guard.ts` 加载 context-mode adapter 与 skills（`context-mode-filter.ts` 过滤冲突锚点） |
 | `extensions/hashline.ts` | 部署 `config/pi-hashline-edit.json` 后加载 pi-hashline-edit 覆盖 `read`/`edit`（`hashline-config.ts`） |
 | `extensions/permission-system.ts` | 直接加载仓库内 `pi-permission-system/src/index.ts` |
 | `extensions/lsp.ts` | 物化包内 typescript-language-server、部署 LSP 配置并加载 pi-lsp（`lsp-config.ts`） |
@@ -35,8 +35,8 @@
 ## 全局硬规则
 
 - 系统提示词单一所有权：`extensions/system-prompt.ts` 排在 `pi.extensions` 最后，是唯一允许在 `before_agent_start` 返回 `systemPrompt` 的扩展；其他扩展只能经 `systemPromptOptions.sections.<name>` 贡献分区，第三方扩展必须经 `extensions/` wrapper + `prompt-guard.ts` 加载，新增工具必须在 `config/prompt/tools.jsonc` 登记。宿主传入 `customPrompt`（子 Agent / SYSTEM.md）时不覆盖其开场白；子 Agent 有独立提示词画像 `config/prompt/subagent/`（preamble + 内置 Agent），判定规则与路由 scope/`subagentExclude` 细则见 `extensions/AGENTS.md`。
-- 加载顺序：`permission-system` → `subagents` → `pi-ssh` → `pi-chrome` → `pi-tool-search` → `system-prompt`（最后）；`pi-hermes-memory` 必须排在 `pi-compaction` 之前（`session_before_compact` 取消语义），`pi-compaction` 在 `pi-minimal-footer` 与 `context-view` 之前。
+- 加载顺序：`pi-overflow/capture.ts` 必须排在最前（先于 RTK 等结果改写扩展暂存原文），`permission-system` → `subagents` → `pi-ssh` → `pi-chrome` → `pi-tool-search` → `pi-overflow/index.ts` → `system-prompt`（最后）；`pi-hermes-memory` 必须排在 `pi-compaction` 之前（`session_before_compact` 取消语义），`pi-compaction` 在 `pi-minimal-footer` 与 `context-view` 之前。
-- 依赖：外部依赖在根 `package.json` 用精确版本，自维护扩展一律 `file:` 本地依赖并更新锁文件；Pi 核心包只作 optional host peer，根 runtime deps 不含 Pi runtime，任何目录不得再装一套 Pi runtime；各子包用于测试的 `@earendil-works/*` devDependencies 一律写 `"latest"`，经 `scripts/update-pi-devdeps.sh` 统一刷新；原生构建依赖只按锁定版本进根 `allowScripts`（当前仅 `better-sqlite3@12.11.1`；pi-hermes-memory 的 `better-sqlite3@13` 用包内 N-API 预编译，不新增放行）。不要用仓库级 `.pi/settings.json` 代替组合包依赖。
+- 依赖：外部依赖在根 `package.json` 用精确版本，自维护扩展一律 `file:` 本地依赖并更新锁文件；Pi 核心包只作 optional host peer，根 runtime deps 不含 Pi runtime，任何目录不得再装一套 Pi runtime；各子包用于测试的 `@earendil-works/*` devDependencies 一律写 `"latest"`，经 `scripts/update-pi-devdeps.sh` 统一刷新；原生构建依赖只按锁定版本进根 `allowScripts`（当前无放行项；pi-hermes-memory 的 `better-sqlite3@13` 用包内 N-API 预编译，不经根 `allowScripts`）。不要用仓库级 `.pi/settings.json` 代替组合包依赖。根 `files` 对子包按运行时子路径白名单列出（源码、package.json、README/LICENSE/CHANGELOG/UPSTREAM），不得整目录列入，以免打入嵌套 node_modules、dist 和测试；由 `tests/package-files.test.ts` 守护。
 - 配置：`config/` 下全部文件是权威源、必须提交；不得读写 `settings.json` 自定义段、agent 目录独立配置文件或以环境变量作持久默认值；`config/` 只放机器无关值，路径类默认值在代码中运行时计算。`<agentDir>/my-pi/` 只放可丢弃运行时状态（日志、缓存），连同旧 `extensions/pi-permission-system/` 残留、构建产物与依赖目录都不提交；用户数据（记忆、会话、`auth.json` 等）不属于组合包配置，也不得放入该目录。
 - 组合包不加载任何 MCP adapter/SDK：Exa 与 CodeGraph 均用自有直连客户端。
 - 导入/同步上游：只从官方 Git 仓库的明确 tag/commit 取源码快照，禁止 npm tarball/缓存/`node_modules`/`dist/` 等预编译产物作导入源；保留上游 LICENSE、作者、测试与来源说明，移除嵌套 `.git`、`.pi` 状态、`node_modules` 与构建产物；同步前先核对本地改动再按明确范围移植，不得整体覆盖。各目录的维护约定见 `AGENTS.md`，每个自维护目录必须有中文 `UPSTREAM.md` 记录来源，上游原文文件不翻译不改写。

README.md

@@ -4,7 +4,6 @@
 
 - `@ff-labs/pi-fff@0.10.5`：使用官方 `override` 模式接管 `find`、`grep` 和 `multi_grep`。
 - 本仓库维护的 `pi-rtk-optimizer`：只处理 Bash/read 等输出压缩，默认不改写命令，也不处理搜索工具结果。
-- `context-mode@1.0.169`：通过 Pi adapter 提供 `ctx_*` 工具、上下文隔离和会话连续性。
 - 本仓库自维护的 `pi-web-search`：对模型只暴露 `web_search` / `web_fetch` 两个工具；`web_search` 由当前主模型在 Tavily / Exa / Keenable 间路由，`web_fetch` 固定走 Exa，三家均以纯 `fetch` 直连（Exa 为 `mcp.exa.ai` 的 JSON-RPC，不经 MCP adapter）。
 - `extensions/codegraph.ts`：注册原生 `codegraph_explore` 工具，按索引根惰性启动常驻 `codegraph serve --mcp` 子进程，由组合包自有 stdio JSON-RPC 客户端直连（无 MCP adapter/SDK）；serve 的文件监听自动同步索引，无结果缓存；未初始化项目经确认后才 `codegraph init`，拒绝记入会话不再重复询问。
 - `extensions/context-view.ts`：通过 `/context` 只读显示 `pi-compaction` 的 effective context，不建立第二套上下文口径。
@@ -12,11 +11,12 @@
 - `pi-hermes-memory/`：从官方 `pi-hermes-memory v0.9.10` 源码导入的持久记忆扩展，提供全局/项目/用户/失败四类记忆、SQLite FTS5 会话与记忆检索、`SKILL.md` 过程记忆、correction 检测与自动 consolidation。
 - 本仓库维护的 `pi-minimal-footer`：基于 `@ogulcancelik/pi-minimal-footer@0.1.10`，用紧凑的上下文仪表和订阅用量条替换默认 footer。
 - 本仓库维护的 `pi-notify`：基于 `@smoose/pi-notify@0.1.1`，使用 Kitty OSC 99 在 Pi 完全 settled 或权限系统即将请求人工确认时通知，并支持点击精确聚焦来源 Kitty 窗口。
-- 本仓库维护的 `pi-compaction`：以增量投影统一确定性工具结果压缩、后台语义摘要、effective token 计量、`context_tree_query`、portable checkpoint 与 Codex native compaction。
+- 本仓库维护的 `pi-compaction`：压力触发时以 Pi `context_edit` 批量遮蔽旧工具结果、空闲时按阈值触发整段摘要、effective token 计量、`context_tree_query`、portable checkpoint 与 Codex native compaction。
+- 本仓库自维护的 `pi-overflow`：统一的新工具结果展示截断——超阈值原文落盘、向模型展示「头 + 信号行 + 尾」的有界视图，并提供 `overflow_query` 按 id 搜索、行区间或 JSON 路径取回原文。
 - 本仓库维护的 `pi-ask-user`：提供模型主动调用的 `ask_user_question`；TUI 中保留单题、多题、选择、文本、自定义回答与提交前复核，Paseo 等兼容 RPC 客户端通过标准 Pi extension UI 逐题交互。
 - 本仓库自维护的 `pi-worktree`：纯斜杠命令（`/wt` / `/worktree`）git worktree 管理，`/wt new` 一键拆分、`/wt merge` 一键合并回来源分支（默认 ff-only，可选 `--squash`），通过 `SessionManager.forkFrom` + `switchSession` 切换会话 cwd；登记状态在 `<agentDir>/pi-worktree/state.json`，不注册模型工具、不 push。
 - 本仓库维护的 `pi-tool-search`：从完整工具定义生成并缓存经过校验的工作流分组，默认以最多 5 个动态组、28 个动态工具的 LRU 策略加载原始完整 schema，并支持原子批量激活。
-- 本仓库维护的 `pi-chrome`：从上游 `0.15.46` 源码导入，通过显式会话授权接入用户当前已登录的 Chrome Profile；大体积页面快照写入 Session 临时文件并交给 Context Mode 隔离分析，21 个 `chrome_*` 工具仍由 Tool Search 按工作流加载。
+- 本仓库维护的 `pi-chrome`：从上游 `0.15.46` 源码导入，通过显式会话授权接入用户当前已登录的 Chrome Profile；页面快照 JSON 直接作为工具结果返回，超长时由 pi-overflow 截断并用 `overflow_query` 查询，21 个 `chrome_*` 工具仍由 Tool Search 按工作流加载。
 - 本仓库维护的 `pi-permission-system`：只读取一份全局普通策略与一份独立 Hard Guards，按 pattern 具体度裁决 allow/ask/deny，并整合 Codex Guardian 风格 AutoReview 与人工 `defer` 兜底；自动评审模型完全由配置文件指定，未配置时提示并转人工审批；工具显隐仍只由 Tool Search 管理。
 - 本仓库维护的 `pi-subagents`：从 `@tintinweb/pi-subagents@0.18.0` 源码导入并默认加载，提供隔离子 Agent、后台执行、steering、resume、嵌套代理与自定义 Agent；通过 mandatory permission wrapper 接入权限 child lifecycle、根 authority `ask` 转发和根 transcript auto-review 身份桥接，其 3 个编排工具由 Tool Search 的 `subagents` 组按需加载。子 Agent 使用独立提示词画像 `config/prompt/subagent/`（自有 preamble 与内置 Agent 定义），不嵌入父会话提示词。
 - 本仓库维护的 `pi-ssh`：通过纯 Node `ssh2` 持久连接、显式 `ssh_cd` 远端工作区切换、SFTP 与自适应有界搜索提供独立的远端工具；Agent 只在用户明确指定已导入主机及具体任务后发起受 AutoReview 复核的连接。
@@ -141,7 +141,7 @@ CodeGraph 的更新或遥测设置。
 
 在线搜索默认按任务只选一个服务：关键词不确定、新闻/近期话题或需要较多候选来源时优先 Tavily；官方/API 文档、版本信息、论文和高确定性事实确认优先 Exa；中文政策、政府公告、中文网页、站点限定和日期筛选优先 Keenable。只有来源质量或不确定性确实需要时才跨服务复核，避免默认把三套结果全部塞进上下文。
 
-日志、测试/构建输出、工作区内大文件的探索/分析/总结以及不可预测的大输出优先交给 Context Mode。只有确实需要精确源码或准备修改时，才用带 `offset/limit` 的 Hashline `read` 读取最小区域并取得 `LINE#HASH` 锚点，随后用锚定 `edit` 修改；成功编辑返回的新锚点可继续复用，只有锚点过期或下一目标区域尚未展示时才重新读取。已有工具结果足够时停止检索，避免对同一问题依次重复调用 CodeGraph、FFF 和 `read`。
+日志、测试/构建输出以及不可预测的大输出由 pi-overflow 截断展示，原文可用 `overflow_query` 按 id 检索。只有确实需要精确源码或准备修改时，才用带 `offset/limit` 的 Hashline `read` 读取最小区域并取得 `LINE#HASH` 锚点，随后用锚定 `edit` 修改；成功编辑返回的新锚点可继续复用，只有锚点过期或下一目标区域尚未展示时才重新读取。已有工具结果足够时停止检索，避免对同一问题依次重复调用 CodeGraph、FFF 和 `read`。
 
 使用 `/dump-system-prompt` 可将当前扩展所见的有效提示词写入当前项目的 `.pi-debug/effective-system-prompt.md`。`.pi-debug/` 属于本地诊断输出，默认不提交。
 
@@ -149,7 +149,7 @@ CodeGraph 的更新或遥测设置。
 
 `extensions/system-prompt.ts` 是组合包唯一的系统提示词出口，排在 `pi.extensions` 最后加载。提示词内容全部由 `config/prompt/` 管理：`preamble.md` 替换 Pi 默认开场白（其余 `<tools>`、`<rules>`、`<skills>`、`<project_context>`、`<cwd>` 等分区及顺序原样保留），`tools.jsonc` 覆盖各工具的简介与规则，`routing.jsonc` 渲染 `<tool_routing>` 分区，`sections.jsonc` 决定各扩展贡献分区的顺序。
 
-其他扩展不直接改写提示词：pi-chrome、pi-ssh、pi-hermes-memory 等通过 `systemPromptOptions.sections.<name>` 贡献结构化分区，pi-rtk-optimizer 通过 `promptGuidelines` 补充规则；第三方扩展（FFF、Hashline、LSP、Context Mode）经 `extensions/` 下的包装文件与 `prompt-guard.ts` 代理加载，拦截直接的提示词改写并按 `tools.jsonc` 应用工具说明覆盖。
+其他扩展不直接改写提示词：pi-chrome、pi-ssh、pi-hermes-memory 等通过 `systemPromptOptions.sections.<name>` 贡献结构化分区，pi-rtk-optimizer 通过 `promptGuidelines` 补充规则；第三方扩展（FFF、Hashline、LSP）经 `extensions/` 下的包装文件与 `prompt-guard.ts` 代理加载，拦截直接的提示词改写并按 `tools.jsonc` 应用工具说明覆盖。
 
 `config/prompt/` 每轮重新读取（按 mtime 缓存），修改已安装组合包内的文件后下一轮即生效；配置解析失败时警告一次并保持 Pi 默认提示词，不阻止会话。该目录由组合包直接维护并随组合包发布；对已安装副本的本地修改不做保护，`pi update` 会直接覆盖；要长期自定义请直接编辑仓库中的 `config/prompt/`。
 
@@ -159,7 +159,7 @@ CodeGraph 的更新或遥测设置。
 
 授权后注册的 21 个 `chrome_*` 工具不会常驻提示词；Tool Search 将它们分为 `chrome-navigation`、`chrome-interaction` 和 `chrome-debugging`。普通网页资料发现仍使用 Tavily/Exa/Keenable，只有需要读取登录态页面或实际操作当前 Chrome 时才加载 Chrome 组。
 
-`chrome_snapshot`、`chrome_find` 和带 `includeSnapshot=true` 的交互不会再把完整页面结构直接塞进模型上下文：完整 JSON 只写入当前项目 `.pi/chrome-context/` 下按 Session/目标隔离的临时文件，工具结果仅返回相对路径和小型摘要。模型应立即用 `ctx_execute_file` 从该文件提取下一步所需的 uid、文本或状态，不能用 `read` / `cat` 重新读入全文。每个目标只保留最新快照；页面操作后旧快照失效，Session 结束、`/chrome revoke` 或授权过期时清理。
+`chrome_snapshot` 和带 `includeSnapshot=true` 的交互把页面快照 JSON 直接作为工具结果返回，`details` 只保留元素数、匹配数等小型元数据；`chrome_find` 只返回紧凑的排序匹配。快照超长时结果会带 `[pi-overflow id=…]` 标记，模型用 `overflow_query` 的 `jsonPath`（如 `elements[0].uid`）或 `search` 提取所需的 uid、文本或状态，而不是重复抓取快照。页面交互后旧快照即失效，需要重新抓取；早期版本写入 Session 临时文件的快照机制已移除。
 
 ### 按需加载工具 schema
 
@@ -167,7 +167,7 @@ CodeGraph 的更新或遥测设置。
 
 本地源码直接使用 `@earendil-works/pi-coding-agent`、当前 `typebox`、`ModelRegistry.complete()` 与 `setActiveTools()`，不安装、别名映射或加载旧 `@mariozechner` runtime，也不使用 provider payload 改写、代理执行或隐藏 `sendMessage` 循环。
 
-新 session 默认常驻 Pi 核心 `read`、`write`、`edit`、`bash`、`grep`、`find`，以及 `codegraph_explore`、`lsp_diagnostics`、`ctx_execute`、`ctx_execute_file`、`ctx_batch_execute` 和 `tool_search`；Windows 上额外把内置 `powershell` 与 `bash` 作为同类本地 Shell 核心工具固定启用，非 Windows 平台则从目录中过滤它。组合包在 `pi-tool-search/extensions/bundle-groups.ts` 中为自身暴露的工具预置权威分组，包括本地文件导航、SSH 远端文件/命令、Chrome 导航/交互/调试、用户交互、CodeGraph/LSP、Tavily、Exa、Keenable、Context Mode 执行/知识库/观测/管理、Memory 查询/维护与 Skill 管理。不可用、已固定或尚未授权而未注册的可选工具会自动从动态组中滤除。
+新 session 默认常驻 Pi 核心 `read`、`write`、`edit`、`bash`、`grep`、`find`，以及 `codegraph_explore`、`lsp_diagnostics`、六个记忆工具、`context_tree_query`、`overflow_query` 和 `tool_search`；Windows 上额外把内置 `powershell` 与 `bash` 作为同类本地 Shell 核心工具固定启用，非 Windows 平台则从目录中过滤它。组合包在 `pi-tool-search/extensions/bundle-groups.ts` 中为自身暴露的工具预置权威分组，包括本地文件导航、SSH 远端文件/命令、Chrome 导航/交互/调试、用户交互、CodeGraph/LSP、Tavily、Exa、Keenable、Memory 查询/维护与 Skill 管理。不可用、已固定或尚未授权而未注册的可选工具会自动从动态组中滤除。
 
 标准组合包的全部隐藏工具都能命中预置目录，因此首次使用不调用模型、不生成用户缓存，也不把完整隐藏 schema 发送给 provider。只有用户另外安装了未识别工具，并且当前调用不是精确加载已有分组时，才可使用当前已认证模型补充目录；精确已有分组始终立即按当前确定性目录激活，不等待无关未知工具的模型增强。模型结果必须保留组合包预置分组，否则直接拒绝。有效增强缓存以 `0600` 写到 agent 目录的 `my-pi/tool-search/catalog-v1.json`，调用用量计入工具结果。用户 `groupOverrides` 的优先级高于预置目录；`/tool-search-rebuild` 会恢复预置目录，`/tool-search-status` 显示目录来源、容量、LRU 顺序和最近淘汰原因。
 
@@ -180,16 +180,14 @@ Tool Search 的配置文件是已安装组合包内的 `config/tool-search.json`
   "alwaysEnabled": [
     "codegraph_explore",
     "lsp_diagnostics",
-    "ctx_execute",
-    "ctx_execute_file",
-    "ctx_batch_execute",
     "memory_search",
     "session_search",
     "memory_add",
     "memory_replace",
     "memory_remove",
     "skill_manage",
-    "context_tree_query"
+    "context_tree_query",
+    "overflow_query"
   ],
   "showToolSearchFooterStatus": false,
   "maxActiveGroups": 5,
@@ -205,7 +203,7 @@ Tool Search 的配置文件是已安装组合包内的 `config/tool-search.json`
 
 组合包固定加载 `pi-hashline-edit@0.8.3`，以带 `LINE#HASH` 行锚点的实现覆盖 Pi 内置 `read` 和 `edit`。`extensions/hashline.ts` 在加载扩展前把 `config/pi-hashline-edit.json` 同步到 `~/.pi/agent/hashline.json`：`hashLength` 保持最小的 `2`，`grep` 强制关闭以避免与 FFF 冲突，`replaceText` 关闭以要求修改使用可验证锚点。
 
-Hashline 不负责压缩大文件；每行锚点本身还会增加少量 token。上下文节省来自先用 CodeGraph/LSP 定位、再用 Context Mode 隔离分析，最后只对真正准备修改的区域执行小范围 Hashline `read`。因此不应为了探索、计数、比较或总结而直接读取完整大文件。
+Hashline 不负责压缩大文件；每行锚点本身还会增加少量 token。上下文节省来自先用 CodeGraph/LSP 定位，最后只对真正准备修改的区域执行小范围 Hashline `read`。因此不应为了探索、计数、比较或总结而直接读取完整大文件。
 
 ### 搜索归 FFF
 
@@ -220,14 +218,12 @@ RTK 保留 Bash、read、build、test、lint 和 Git 输出压缩。默认 `read
 仍关闭；可用 `/rtk` 查看或调整。旧机器已有的 RTK 配置若没有
 `commandRewritingEnabled` 字段，也会按 `false` 归一化，不需要迁移配置。
 
-### 大输出归 Context Mode
+### 大输出归 pi-overflow
 
-Context Mode 直接加载 npm 包内置的 Pi adapter 和 skills，不需要额外写入
+新工具结果的展示截断统一由 `pi-overflow` 负责：超阈值的原文落盘、模型只看到
-`~/.pi/agent/mcp.json`，也不需要全局安装 `context-mode` 命令。它提供 `ctx_*` 工具，
+「头 + 信号行 + 尾」的有界视图，需要原文时用 `overflow_query` 检索（详见下文
-将批量读取、命令研究和网页内容先在隔离进程中处理，再把收敛后的结果送入上下文。
+「Pi Overflow」小节）。FFF 仍负责精确字面搜索；RTK 继续对 Bash、build、test、lint
-
+和 Git 输出做语义压缩，其兜底硬截断已关闭。
-FFF 仍负责精确字面搜索；RTK 继续压缩未走 Context Mode 的 Bash、build、test、lint
-和 Git 输出。Context Mode 不替换这两个扩展的现有配置。
 
 ### SSH 服务器配置
 
@@ -366,23 +362,31 @@ compaction 和排队 follow-up 完成前不会提前通知；同时监听 `pi-pe
 `permissions:ui_prompt`，仅在权限系统即将显示人工确认界面时立即通知，策略或 AutoReview 已直接完成的
 决定不会通知。Kitty 后端使用 Base64 OSC 99、每个 Pi Session 独立的稳定通知 ID、显式 `a=focus`，
 并默认设置 `o=always`：即使来源 Kitty window/pane 正在接收键盘输入也会显示通知，点击通知由 Kitty
-返回原始来源；仍可通过环境变量改用 `unfocused` 或 `invisible`。
+返回原始来源；仍可通过 `visibility` 改用 `unfocused` 或 `invisible`。
 
 默认忽略不足 3 秒的短任务，但权限确认通知不受该时长门槛影响。`/notify on|off|test|status` 可控制
-当前 Session 的两类通知；持久默认通过 `PI_NOTIFY_ENABLED`、`PI_NOTIFY_MIN_SECONDS`、
+当前 Session 的两类通知；持久默认通过组合包 `config/pi-notify.json` 的 `enabled`、`minSeconds`、
-`PI_NOTIFY_VISIBILITY`、`PI_NOTIFY_ACTION`、`PI_NOTIFY_MESSAGE_SOURCE` 等环境变量配置。
+`visibility`、`action`、`messageSource` 等字段配置。`messageSource: "none"`
-`PI_NOTIFY_MESSAGE_SOURCE=none` 会同时隐藏回复摘要以及权限值和请求者细节。Kitty 位于 tmux 内时需要
+会同时隐藏回复摘要以及权限值和请求者细节。Kitty 位于 tmux 内时需要
 `set -g allow-passthrough all` 并重启 tmux server。完整配置和上游差异见
 [`pi-notify/README.md`](pi-notify/README.md)。
 
 ### Pi Compaction
 
-组合包只加载仓库内维护的 `pi-compaction/`。它以增量 Segment Index 管理已经进入 Session 的历史：已闭合工具结果先同步替换为有界的确定性证据 stub，大于阈值的结果再由单 Session 有界后台 worker 生成语义摘要；Provider 请求不会等待后台摘要。原始工具证据不重复写入 custom entry，可通过 `context_tree_query` 使用 source key 或 tool call ID 有界恢复。
+组合包只加载仓库内维护的 `pi-compaction/`。它管理已经进入 Session 的历史投影：上下文压力达到阈值时，在 `turn_end` 批量提交 Pi 原生 `context_edit`，把最近 `keepRecentTurns` 轮之外的大型工具结果遮蔽为指向 `context_tree_query` 的单行占位符；遮蔽成批发生、受 trigger/target 回差与 `minClearTokens` 约束。原始工具结果保留在 Session 中（UI、导出不变），可通过 `context_tree_query` 按 tool call ID 有界恢复。
 
-普通 Provider、阈值判断、Codex native compaction、`/context`、`/compaction-status` 和 minimal footer 都读取同一个 effective projection snapshot。Pi Core 的 raw threshold 只负责唤醒：只有 effective token 下界达到默认 90% 才执行 threshold compaction；manual 与 overflow 始终允许继续并在 native 失败时回退 Pi 默认 compaction。Codex native API 直接消费 effective ResponseItems，同时保存有界 portable summary 供模型或 Provider 切换。
+普通 Provider、阈值判断、Codex native compaction、`/context`、`/compaction-status` 和 minimal footer 都读取同一个 effective projection snapshot。整段摘要只在空闲 `agent_settled` 且达到绝对 token 阈值（默认 400K，受窗口比例约束，272K 窗口时约为 90%）时经 `ctx.compact` 触发一次；Pi 自身的阈值压缩作为最终安全网始终有效，`session_before_compact` 不再按扩展自身阈值取消。Codex native API 直接消费 effective ResponseItems，同时保存有界 portable summary 供模型或 Provider 切换。
 
-唯一配置位于已安装组合包的 `config/pi-compaction.json`，默认 soft/hard/target 比例为 72%/90%/55%，扩展只在内存中归一化且不写回配置。`/compaction-status` 显示 raw/effective/reclaimed token、置信度、segment 和后台任务。完整配置、Session version 1 格式和双上游来源见 [`pi-compaction/README.md`](pi-compaction/README.md) 与 [`pi-compaction/UPSTREAM.md`](pi-compaction/UPSTREAM.md)。
+唯一配置位于已安装组合包的 `config/pi-compaction.json`：`masking` 段为遮蔽触发/目标阈值、`keepRecentTurns`、`minEntryTokens` 与 `excludeTools`，`summary` 段为整段摘要阈值与 thinking 档位，扩展只在内存中归一化且不写回配置。`/compaction-status` 显示 raw/effective/reclaimed token、置信度、segment 和已遮蔽结果数。完整配置、Session version 1 格式和双上游来源见 [`pi-compaction/README.md`](pi-compaction/README.md) 与 [`pi-compaction/UPSTREAM.md`](pi-compaction/UPSTREAM.md)。
 
 原生远程压缩同时支持 `openai-codex` + `openai-codex-responses` 和 `sub-api` + `openai-responses`，使用触发时当前会话模型。SubAPI 保留现有模型配置，从 `baseUrl` 调用公共 `/responses` 并使用 `compaction_trigger` 流式协议；优先保留非空显式 `Authorization`，否则使用 API key，也支持仅凭非空 `cf-aig-authorization` 的代理认证。缺失或空白凭据在请求前拒绝。SubAPI 不要求客户端解析 ChatGPT account ID，也不调用独立 `/responses/compact`；Codex 的 JWT 要求保持不变。网关部署须支持 Sub2API 原生压缩 v2；请求失败沿用 Pi 回退，切换模型或 Provider 时使用 portable summary。
 
-本实现不读取或迁移旧 `contextPrune`、`context-prune-*` 或旧 Codex native checkpoint；切换后应创建新 Session。它与 Context Mode 职责互补：Context Mode 防止大型原始内容进入会话，Pi Compaction 管理已经进入会话的长期历史。
+本实现不读取或迁移旧 `contextPrune`、`context-prune-*` 或旧 Codex native checkpoint；切换后应创建新 Session。它与 pi-overflow 职责互补：pi-overflow 在新工具结果进入会话时即截断展示，Pi Compaction 管理已经进入会话的长期历史。
+
+### Pi Overflow
+
+组合包自维护的 `pi-overflow/` 统一负责「每个新工具结果给模型看多少」。它有两个入口：`capture.ts` 排在 `pi.extensions` 最前，只把原始结果文本暂存到进程内 Map（容量 64），绝不修改结果；`index.ts` 排在 `system-prompt.ts` 之前、所有其他扩展之后，此时 RTK 等 handler 已经完成变换——它用最终可见文本判断是否超阈值，落盘的原文取 bash 的 `details.fullOutputPath` 文件、暂存原文或当前文本三者中的最完整版本，视图与其中的行号则一律基于落盘原文生成，保证与 `overflow_query` 读取的内容一致。
+
+当最终文本仍超过 `thresholdChars`（默认 24000）时，模型看到的是确定性的「头部 40 行 + 中段匹配 error/warn 等信号行（至多 40 条）+ 尾部 120 行」视图，并写入 `details.overflow.id`；当上游已把结果压短但原文超阈值时，只追加一行 `[pi-overflow id=…]` 恢复标记。原文以 `sha1(sessionId+toolCallId)` 前 12 位 hex 为 id 存入 `<agentDir>/my-pi/pi-overflow/`（目录 `700`、文件 `600`、原子写入），session 启动时按 `retentionDays`/`maxStoreBytes` 清理。模型可用 `overflow_query` 按 id 正则搜索（带上下文合并）、按行区间或 `jsonPath` 读取；`read`/`edit`/`write`/`overflow_query`/`ask_user_question` 与嵌套调用永远豁免。配置为组合包内 `config/pi-overflow.json`，详见 [`pi-overflow/README.md`](pi-overflow/README.md)。
+
+RTK 的兜底截断与 pi-chrome 的 30K 字符截断已收编进本扩展：`config/rtk-optimizer.json` 中 `outputCompaction.truncate.enabled` 默认关闭，chrome_* 工具返回完整文本；pi-web-search 的 `web_fetch` 默认 `max_characters` 提高到 20000（上限 100000），`web_search` 的 `max_results` 约束在 1–20。

config/pi-compaction.json

@@ -1,20 +1,20 @@
 {
   "enabled": true,
-  "softRatio": 0.72,
+  "masking": {
-  "hardRatio": 0.9,
-  "targetRatio": 0.55,
-  "deterministic": {
     "enabled": true,
-    "maxStubTokens": 1024
+    "triggerTokens": 200000,
+    "triggerRatio": 0.6,
+    "targetTokens": 100000,
+    "targetRatio": 0.35,
+    "minClearTokens": 50000,
+    "keepRecentTurns": 10,
+    "minEntryTokens": 256,
+    "excludeTools": ["ask_user_question"]
   },
-  "semantic": {
+  "summary": {
-    "enabled": true,
+    "triggerTokens": 400000,
-    "model": "default",
+    "triggerRatio": 0.9,
-    "thinking": "low",
+    "thinking": "low"
-    "concurrency": 1,
-    "minInputChars": 4000,
-    "maxInputChars": 24000,
-    "maxRetries": 2
   },
   "native": {
     "enabled": true

config/pi-minimal-footer.json

@@ -0,0 +1,4 @@
+{
+  "showCwd": true,
+  "showBranch": true
+}

config/pi-notify.json

@@ -0,0 +1,11 @@
+{
+  "enabled": true,
+  "title": "π - {project}",
+  "body": "{message}",
+  "messageSource": "assistant",
+  "messageMax": 80,
+  "minSeconds": 3,
+  "visibility": "always",
+  "action": "focus",
+  "soundCommand": ""
+}

config/pi-overflow.json

@@ -0,0 +1,13 @@
+{
+  "enabled": true,
+  "thresholdChars": 24000,
+  "headLines": 40,
+  "tailLines": 120,
+  "maxSignalLines": 40,
+  "signalPattern": "\\b(error|errors|fail|failed|failure|warn|warning|exception|traceback|panic|fatal)\\b",
+  "maxLineChars": 500,
+  "exemptTools": ["read", "edit", "write", "overflow_query", "ask_user_question"],
+  "retentionDays": 7,
+  "maxStoreBytes": 1073741824,
+  "maxEntryBytes": 67108864
+}

config/pi-permission-policy.jsonc

@@ -45,13 +45,6 @@
     "lsp_references": "allow",
     "lsp_hover": "allow",
     "lsp_symbols": "allow",
-    "ctx_execute": "allow",
-    "ctx_execute_file": "allow",
-    "ctx_batch_execute": "allow",
-    "ctx_search": "allow",
-    "ctx_index": "allow",
-    "ctx_stats": "allow",
-    "ctx_doctor": "allow",
     "tool_search": "allow",
     "ask_user_question": "allow",
     // Launching, steering or reading a subagent has no side effect itself;
@@ -86,10 +79,12 @@
     // index and never touches the workspace. It is pinned in Tool Search's
     // alwaysEnabled set, so it must not pay an AutoReview round trip per turn.
     "context_tree_query": "allow",
+    // pi-overflow's inspector reads only the extension's own stored output
+    // directory (id-validated); pinned in Tool Search's alwaysEnabled set.
+    "overflow_query": "allow",
 
     // Model-chosen URLs can exfiltrate data — same reason curl/wget ask.
     "web_fetch": "ask",
-    "ctx_fetch_and_index": "ask",
 
     "ssh_connect": "ask",
     "ssh_cd": "ask",

config/prompt/routing.jsonc

@@ -50,20 +50,12 @@
       "text": "Use web_search for current, real-world, or citable web information; provider selection is handled internally. Keep queries focused and use site/published_* filters only when they materially narrow results. Read selected pages with web_fetch instead of repeating similar searches."
     },
     {
-      "whenAll": ["web_fetch", "ctx_fetch_and_index"],
+      "whenAll": ["chrome_snapshot", "overflow_query"],
-      "text": "Read a web page with web_fetch. Use ctx_fetch_and_index only when the content should be indexed into the context-mode knowledge base for repeated ctx_search queries."
+      "text": "chrome_snapshot (and Chrome actions with includeSnapshot=true) return the page snapshot JSON directly. When it is large the result carries a [pi-overflow id=…] marker: extract only the uids, text, or state you need with overflow_query (jsonPath or search) instead of taking another snapshot, and take a fresh snapshot after any page interaction because the previous one is stale."
     },
     {
-      "whenAny": ["ctx_execute", "ctx_execute_file", "ctx_batch_execute"],
+      "whenAny": ["overflow_query"],
-      "text": "Use Context Mode for logs, test/build output, generated data, large-file exploration/analysis/summarization, and any command whose output may be large or unpredictable. Prefer one ctx_batch_execute over several ctx_execute calls for multi-command research, and print only the derived answer needed for the task."
+      "text": "When a tool result starts with or ends with a [pi-overflow id=…] marker, the full original output is stored; inspect it with overflow_query (search, line ranges, or jsonPath) instead of re-running the command or dumping it again."
-    },
-    {
-      "whenAny": ["ctx_execute_file"],
-      "text": "ctx_execute_file analyzes a file and never edits it; make changes with read + edit. It is confined to the current project root: do not call it for absolute paths outside the workspace, ../ traversal, or symlinks that resolve outside the workspace; host permission approval does not bypass this boundary. Use another explicitly authorized tool for those files."
-    },
-    {
-      "whenAll": ["chrome_snapshot", "ctx_execute_file"],
-      "text": "After chrome_snapshot or a Chrome action with includeSnapshot=true, immediately analyze the returned workspace-relative capture path with ctx_execute_file; never use read or cat for the raw JSON. Print only the uids, text, or state needed for the next action, and take a fresh snapshot after any interaction because the previous capture may be stale."
     },
     {
       "whenAny": ["read"],

config/prompt/tools.jsonc

@@ -43,18 +43,6 @@
     "web_search": { "inherit": true },
     "web_fetch": { "inherit": true },
 
-    "ctx_execute": { "inherit": true },
-    "ctx_execute_file": { "inherit": true },
-    "ctx_batch_execute": { "inherit": true },
-    "ctx_index": { "inherit": true },
-    "ctx_search": { "inherit": true },
-    "ctx_fetch_and_index": { "inherit": true },
-    "ctx_stats": { "inherit": true },
-    "ctx_doctor": { "inherit": true },
-    "ctx_insight": { "inherit": true },
-    "ctx_upgrade": { "inherit": true },
-    "ctx_purge": { "inherit": true },
-
     "memory_search": { "inherit": true },
     "session_search": { "inherit": true },
     "memory_add": { "inherit": true },
@@ -62,6 +50,7 @@
     "memory_remove": { "inherit": true },
     "skill_manage": { "inherit": true },
     "context_tree_query": { "inherit": true },
+    "overflow_query": { "inherit": true },
 
     "ask_user_question": { "inherit": true },
 

config/rtk-optimizer.json

@@ -11,7 +11,7 @@
       "enabled": false
     },
     "truncate": {
-      "enabled": true,
+      "enabled": false,
       "maxChars": 12000
     },
     "sourceCodeFilteringEnabled": false,

config/tool-search.json

@@ -2,16 +2,14 @@
   "alwaysEnabled": [
     "codegraph_explore",
     "lsp_diagnostics",
-    "ctx_execute",
-    "ctx_execute_file",
-    "ctx_batch_execute",
     "memory_search",
     "session_search",
     "memory_add",
     "memory_replace",
     "memory_remove",
     "skill_manage",
-    "context_tree_query"
+    "context_tree_query",
+    "overflow_query"
   ],
   "showToolSearchFooterStatus": false,
   "maxActiveGroups": 5,

extensions/AGENTS.md

@@ -27,7 +27,6 @@
 - `fff-override.ts`：强制 FFF 官方 `override` 模式，统一接管 `find`、`grep`、`multi_grep` 和 `@` 补全；FFF 独占本地字面搜索，远端服务器搜索由 `ssh_find` / `ssh_grep` 负责，RTK 不得接管搜索工具（见 `pi-rtk-optimizer/AGENTS.md`）。显式 CLI flag 仍遵循 FFF 官方优先级。
 - `codegraph.ts` + `codegraph-client.ts`：注册原生 `codegraph_explore` 工具，按索引根惰性常驻 `codegraph serve --mcp -p <root>` 子进程，经组合包自有 stdio 逐行 JSON-RPC 客户端直连（不经 shell、不带 pi-mcp-adapter 或 MCP SDK）；serve 进程内置文件监听自动同步索引，无结果缓存。子进程 10 分钟无调用自动关闭、会话结束统一关闭；调用中死亡换新客户端透明重试一次；其启动的上游 daemon（`.codegraph/daemon.sock`）退出后可多驻留 ≤5 分钟，属上游行为，组合包不管理。
 - CodeGraph 是机器级依赖，`install.sh`/`update.sh` 经官方安装器管理 `codegraph` CLI，组合包不内置其运行时。安装、升级和普通启动流程不得执行 `codegraph init`；只有 `codegraph_explore` 在未初始化项目上被调用、且用户在 `ctx.ui.confirm` 中明确确认后才可 `codegraph init` 当前目录，拒绝记入会话级 declined 不再重复询问；init 失败必须如实报告并让模型回退内置工具；不得改动 CodeGraph 更新或遥测设置。CLI 缺失、init 被拒或服务调用失败都不得抛出，必须以文本结果打回并引导模型改用 `find`/`grep`/`read`/LSP。已移除的 `plugin-init.ts`（原 `/plugin_init`）职责已并入本文件的惰性确认，不得恢复。
-- `context-mode.ts` + `context-mode-filter.ts`：经 `prompt-guard.ts` 代理加载上游 `context-mode` Pi adapter 与 skills，提供 `ctx_*` 工具、隔离式大输出处理和会话连续性；Context Mode 负责避免批量读取、命令研究和网页原始内容直接撑大上下文。filter 只丢弃其 context 注入中与本组合包路由重复且冲突的锚点段。
 - `hashline.ts` + `hashline-config.ts`：部署组合包权威 `config/pi-hashline-edit.json` 后加载 `pi-hashline-edit`；默认以 2 字符行哈希覆盖内置 `read` / `edit`，`grep` 与 `replaceText` 均关闭。它保证锚定编辑而不负责压缩，大文件上下文节省必须来自先定位/隔离分析、再做最小范围 `offset/limit` 读取。
 - `permission-system.ts`：直接加载仓库内 `pi-permission-system/src/index.ts`，自身不部署、不播种、不迁移任何配置；运行时直接读取已安装组合包内 `config/pi-permission-policy.jsonc` / `config/pi-permission-hard-guards.jsonc`；日志保留在 `<agentDir>/my-pi/permission-system/logs/`。旧 `extensions/pi-permission-system/` 位置的残留文件被忽略，不迁移也不清理；项目级 policy 不读取。
 - `lsp.ts` + `lsp-config.ts`：TypeScript/JavaScript 后端由根包固定依赖提供，用当前 Node 可执行文件直接启动包内 `typescript-language-server` CLI，不得依赖或调用 VS Code GUI；Kotlin/JDT LS 使用组合包部署的 `kotlin-lsp --stdio` 与 `jdtls` 配置，根包不安装这两个系统可执行文件。

extensions/context-mode-filter.ts

@@ -1,36 +0,0 @@
-/**
- * Pure filter applied after context-mode's `context` handler: its upstream
- * build appends one `{ role: "user", content }` message whose parts are
- * `"\n\n"`-joined and whose first part is a routing anchor that duplicates —
- * and contradicts — the bundle-owned `config/prompt/routing.jsonc`. The anchor
- * part is stripped; remaining parts (active memory, resume snapshot) stay.
- * No other message is touched; when nothing changes the input array is
- * returned unchanged (same reference).
- */
-
-export const CONTEXT_MODE_ANCHOR_PREFIX = "context-mode active. Hierarchy:";
-
-type MessageLike = { role?: unknown; content?: unknown } & Record<string, unknown>;
-
-export function filterContextModeMessages<T extends MessageLike>(
-	before: readonly T[],
-	after: readonly T[],
-): T[] {
-	if (after.length <= before.length) return after as T[];
-	let result: T[] | null = null;
-	for (let i = after.length - 1; i >= before.length; i--) {
-		const message = after[i];
-		if (message === null || typeof message !== "object") continue;
-		if (message.role !== "user" || typeof message.content !== "string") continue;
-		const parts = message.content.split("\n\n");
-		const kept = parts.filter((part) => !part.startsWith(CONTEXT_MODE_ANCHOR_PREFIX));
-		if (kept.length === parts.length) continue;
-		result ??= after.slice();
-		if (kept.length === 0) {
-			result.splice(i, 1);
-		} else {
-			result[i] = { ...message, content: kept.join("\n\n") };
-		}
-	}
-	return result ?? (after as T[]);
-}

extensions/context-mode.ts

@@ -1,18 +0,0 @@
-import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
-// The adapter is a prebuilt .js bundle, not a TypeScript source.
-import piExtension from "../node_modules/context-mode/build/adapters/pi/extension.js";
-
-import { filterContextModeMessages } from "./context-mode-filter.ts";
-import { guardExtensionApi } from "./prompt-guard.ts";
-
-/**
- * Load context-mode through the prompt guard: the bundle keeps sole ownership
- * of the system prompt, so the upstream routing anchor appended to the
- * `context` message list is filtered out (bundle routing lives in
- * `config/prompt/routing.jsonc`), while upstream active-memory/resume parts
- * pass through untouched. The upstream code this filter depends on is pinned
- * by `tests/system-prompt.test.ts` against the exact `context-mode` version.
- */
-export default function bundledContextModeExtension(pi: ExtensionAPI): void {
-	piExtension(guardExtensionApi(pi, { contextFilter: filterContextModeMessages }));
-}

extensions/context-view.ts

@@ -17,7 +17,7 @@ export default function contextView(pi: ExtensionAPI): void {
             `Raw: ${formatNumber(metrics.rawTokens)} / ${formatNumber(metrics.contextWindow)}`,
             `Reclaimed: ${formatNumber(metrics.reclaimedTokens)}`,
             `Segments: ${formatNumber(metrics.segmentCount)}`,
-            `Semantic jobs: ${formatNumber(metrics.semanticJobs)}`,
+            `Masked tool results: ${formatNumber(metrics.maskedResults)}`,
             `Generation: ${formatNumber(metrics.generation)}`,
           ].join("\n")
         : "No pi-compaction projection is available yet. Send one message to build the first snapshot.";

package.json

@@ -10,15 +10,76 @@
   ],
   "files": [
     "extensions",
-    "pi-ask-user",
+    "pi-ask-user/index.ts",
-    "pi-compaction",
+    "pi-ask-user/src",
-    "pi-chrome",
+    "pi-ask-user/package.json",
-    "pi-hermes-memory",
+    "pi-ask-user/README.md",
-    "pi-rtk-optimizer",
+    "pi-ask-user/LICENSE",
-    "pi-minimal-footer",
+    "pi-ask-user/UPSTREAM.md",
-    "pi-subagents",
+    "pi-compaction/index.ts",
-    "pi-notify",
+    "pi-compaction/src",
-    "pi-ssh",
+    "pi-compaction/package.json",
+    "pi-compaction/README.md",
+    "pi-compaction/CHANGELOG.md",
+    "pi-compaction/LICENSES",
+    "pi-compaction/UPSTREAM.md",
+    "pi-chrome/extensions",
+    "pi-chrome/package.json",
+    "pi-chrome/README.md",
+    "pi-chrome/CHANGELOG.md",
+    "pi-chrome/SECURITY.md",
+    "pi-chrome/LICENSE",
+    "pi-chrome/UPSTREAM.md",
+    "pi-hermes-memory/src",
+    "pi-hermes-memory/package.json",
+    "pi-hermes-memory/README.md",
+    "pi-hermes-memory/CHANGELOG.md",
+    "pi-hermes-memory/LICENSE",
+    "pi-hermes-memory/UPSTREAM.md",
+    "pi-rtk-optimizer/index.ts",
+    "pi-rtk-optimizer/src",
+    "pi-rtk-optimizer/package.json",
+    "pi-rtk-optimizer/README.md",
+    "pi-rtk-optimizer/CHANGELOG.md",
+    "pi-rtk-optimizer/LICENSE",
+    "pi-rtk-optimizer/UPSTREAM.md",
+    "pi-minimal-footer/index.ts",
+    "pi-minimal-footer/status.ts",
+    "pi-minimal-footer/settings.ts",
+    "pi-minimal-footer/assets",
+    "pi-minimal-footer/package.json",
+    "pi-minimal-footer/README.md",
+    "pi-minimal-footer/LICENSE",
+    "pi-minimal-footer/UPSTREAM.md",
+    "pi-subagents/src",
+    "pi-subagents/package.json",
+    "pi-subagents/README.md",
+    "pi-subagents/CHANGELOG.md",
+    "pi-subagents/SECURITY.md",
+    "pi-subagents/LICENSE",
+    "pi-subagents/UPSTREAM.md",
+    "pi-overflow/index.ts",
+    "pi-overflow/capture.ts",
+    "pi-overflow/settings.ts",
+    "pi-overflow/src",
+    "pi-overflow/package.json",
+    "pi-overflow/README.md",
+    "pi-overflow/LICENSE",
+    "pi-overflow/UPSTREAM.md",
+    "pi-notify/src",
+    "pi-notify/package.json",
+    "pi-notify/README.md",
+    "pi-notify/LICENSE",
+    "pi-notify/UPSTREAM.md",
+    "pi-ssh/index.ts",
+    "pi-ssh/permission-integration.ts",
+    "pi-ssh/src",
+    "pi-ssh/scripts",
+    "pi-ssh/package.json",
+    "pi-ssh/README.md",
+    "pi-ssh/CHANGELOG.md",
+    "pi-ssh/LICENSE",
+    "pi-ssh/UPSTREAM.md",
     "pi-tool-search/extensions",
     "pi-tool-search/docs",
     "pi-tool-search/package.json",
@@ -26,8 +87,18 @@
     "pi-tool-search/CHANGELOG.md",
     "pi-tool-search/LICENSE",
     "pi-tool-search/UPSTREAM.md",
-    "pi-web-search",
+    "pi-web-search/index.ts",
-    "pi-worktree",
+    "pi-web-search/auth.ts",
+    "pi-web-search/router.ts",
+    "pi-web-search/clients",
+    "pi-web-search/package.json",
+    "pi-web-search/README.md",
+    "pi-web-search/LICENSE",
+    "pi-worktree/index.ts",
+    "pi-worktree/src",
+    "pi-worktree/package.json",
+    "pi-worktree/README.md",
+    "pi-worktree/LICENSE",
     "pi-permission-system/LICENSES",
     "pi-permission-system/docs/auto-review",
     "pi-permission-system/src",
@@ -47,12 +118,12 @@
   ],
   "pi": {
     "extensions": [
+      "./pi-overflow/capture.ts",
       "./extensions/fff-override.ts",
       "./pi-web-search/index.ts",
       "./extensions/codegraph.ts",
       "./pi-rtk-optimizer/index.ts",
       "./extensions/hashline.ts",
-      "./extensions/context-mode.ts",
       "./extensions/lsp.ts",
       "./pi-hermes-memory/src/index.ts",
       "./pi-compaction/index.ts",
@@ -66,10 +137,8 @@
       "./pi-worktree/index.ts",
       "./pi-chrome/extensions/chrome-profile-bridge/index.ts",
       "./pi-tool-search/extensions/index.ts",
+      "./pi-overflow/index.ts",
       "./extensions/system-prompt.ts"
-    ],
-    "skills": [
-      "./node_modules/context-mode/skills"
     ]
   },
   "dependencies": {
@@ -78,13 +147,13 @@
     "@ogulcancelik/pi-minimal-footer": "file:./pi-minimal-footer",
     "@smoose/pi-notify": "file:./pi-notify",
     "@tintinweb/pi-subagents": "file:./pi-subagents",
-    "context-mode": "1.0.169",
     "jiti": "2.7.0",
     "pi-ask-user": "file:./pi-ask-user",
     "pi-chrome": "file:./pi-chrome",
     "pi-compaction": "file:./pi-compaction",
     "pi-hashline-edit": "0.8.3",
     "pi-hermes-memory": "file:./pi-hermes-memory",
+    "pi-overflow": "file:./pi-overflow",
     "pi-lsp": "0.1.7",
     "pi-ssh": "file:./pi-ssh",
     "pi-tool-search": "file:./pi-tool-search",
@@ -117,8 +186,5 @@
   },
   "engines": {
     "node": ">=22.19.0"
-  },
-  "allowScripts": {
-    "better-sqlite3@12.11.1": true
   }
 }

pi-chrome/AGENTS.md

@@ -1,9 +1,8 @@
 # my-pi 维护约定（pi-chrome）
 
-- 扩展入口是 `pi-chrome/extensions/chrome-profile-bridge/index.ts`，Companion MV3 扩展位于 `browser-extension/`，Session 快照存储位于 `src/session-snapshot-store.ts`，测试保留在 `test/` 与 `test-suite/`，来源记录是 `UPSTREAM.md`。
+- 扩展入口是 `pi-chrome/extensions/chrome-profile-bridge/index.ts`，Companion MV3 扩展位于 `browser-extension/`，测试保留在 `test/` 与 `test-suite/`，来源记录是 `UPSTREAM.md`。
 - 同步上游必须从官方 `tianrendong/pi-chrome` 的明确 tag/commit 移植并保留 README、CHANGELOG、SECURITY、文档、测试、作者和 LICENSE；不得导入嵌套 `.git`、`node_modules` 或构建产物。
 - 根包必须通过 `file:./pi-chrome` 提供依赖并直接加载仓库源码，不得同时安装或加载 npm 预编译入口。
 - 控制默认锁定，每个 Pi session 必须显式 `/chrome authorize`；未授权时不注册 Chrome 工具。显式授权后出现的 21 个 `chrome_*` 工具必须由 Tool Search 分入 `chrome-navigation`、`chrome-interaction`、`chrome-debugging`，不得落入 unknown-tools 模型增强路径。授权撤销或过期时必须停用全部 21 个工具，包括 `chrome_find` / `chrome_inspect`；Tool Search 重新激活 schema 不能替代每次执行时的授权校验。
-- 完整 Chrome snapshot 不得进入普通工具文本或 `details`；只能原子写入当前项目 `.pi/chrome-context/<session>/<target>/snapshot.json`，使用哈希目录、POSIX `700` 目录 / `600` 文件，并只返回工作区相对路径和有界元数据。快照不得跨 Session 持久化，Session 启动/关闭、撤销和授权过期必须清理。
+- `chrome_snapshot` 与带 `includeSnapshot=true` 的动作把快照 JSON 直接放进工具结果文本返回；超长结果的展示截断由 pi-overflow 统一负责（原文落盘，模型用 `overflow_query` 的 jsonPath/search 提取）。`details` 只放元素数、匹配数等小型元数据，不得把完整快照写进 `details`（会进入会话文件），也不再写 Session 临时快照文件。`chrome_find` 只返回有界匹配摘要，不保存文件。
-- `chrome_snapshot` 和带 `includeSnapshot=true` 的动作必须引导模型立即使用 `ctx_execute_file`，不得使用 `read` / `cat` 把原始 JSON 重新塞入上下文；任何可能改变页面或目标的成功操作必须使旧快照失效，`chrome_find` 只可额外返回有界匹配摘要。
 - 修改 bridge、Companion 扩展、授权 lifecycle、快照存储或加载入口后至少运行包内 `npm test`、根 Tool Routing/Tool Search/扩展联合加载测试、`npm pack --dry-run --json` 和实际 packed tarball 隔离安装验证；未经明确要求不得连接或控制真实 Chrome。

pi-chrome/README.md

@@ -6,7 +6,7 @@
 
 ```text
 You:    "Find my open GitHub PR tab, summarize review state, and screenshot failing CI."
-Agent:  chrome_tab(list) → chrome_snapshot(...) → ctx_execute_file(capture) → chrome_screenshot(...)
+Agent:  chrome_tab(list) → chrome_snapshot(...) → overflow_query(jsonPath) → chrome_screenshot(...)
         ✓ 3 reviewers, 1 change requested, CI red on iOS. Saved → .pi/chrome-screenshots/ci.png
 You:    [keeps coding — agent never asked you to log in]
 ```
@@ -74,17 +74,11 @@ Second doctor run should show all checks passing.
 
 Tool parameters and gotchas are documented inline in Pi.
 
-### Session-local snapshots and Context Mode
+### Snapshots and pi-overflow
 
-In this maintained build, `chrome_snapshot`, `chrome_find`, and actions with `includeSnapshot=true` keep the full structured page snapshot out of the normal tool result. They atomically save only the latest capture for each target under:
+In this maintained build, `chrome_snapshot` and actions with `includeSnapshot=true` return the full structured page snapshot as JSON directly in the tool result. `chrome_find` returns only up to 20 compact ranked matches.
 
-```text
+Large snapshots are bounded by the bundle's `pi-overflow` extension: the result keeps a bounded view and the full original is stored, marked with `[pi-overflow id=…]`. Extract the uids, text, or state you need with `overflow_query` — a `jsonPath` such as `elements[0].uid` or `summary.focused`, or a regex `search` — instead of taking another snapshot. Any page interaction makes earlier snapshots stale, so take a fresh one (or pass `includeSnapshot=true` on the action) when the page has changed.
-.pi/chrome-context/<session-hash>/<target-hash>/snapshot.json
-```
-
-The tool returns a workspace-relative path plus bounded metadata. Analyze that path immediately with `ctx_execute_file` and print only the uids, text, or state needed for the next browser action. Do **not** use `read` or `cat` on the capture, because that would put the raw page back into model context.
-
-Captures are deliberately ephemeral: a successful page interaction invalidates the old target capture, and session shutdown, `/chrome revoke`, or authorization expiry removes the session files. `chrome_find` still returns up to 20 compact ranked matches for direct use while storing the full capture privately.
 
 ---
 

pi-chrome/UPSTREAM.md

@@ -15,10 +15,9 @@
 
 导入后的本地改动：
 
-- Chrome 快照以原子方式写入当前项目 `.pi/chrome-context/` 目录下的 owner-only、Session 临时文件，而不是把原始页面结构返回给模型上下文。
+- Chrome 快照与 `includeSnapshot=true` 的动作结果把快照 JSON 直接放入工具文本返回，`details` 只保留元素数/匹配数等小型元数据；展示截断由组合包 pi-overflow 统一负责（原文落盘 + `overflow_query` 查询）。`chrome_find` 只返回紧凑排序匹配，不保存文件。早期版本曾把快照原子写入 `.pi/chrome-context/` 临时文件并引导 `ctx_execute_file` 分析；该 Session 临时文件机制已随 context-mode 一起移除，各项目中残留的 `.pi/chrome-context/` 目录不再创建，可手动删除。
-- 每个 Session/目标只保留最新快照；浏览器交互会使陈旧捕获失效，Session 关闭时移除该 Session 目录。
-- `chrome_find` 返回紧凑的排序匹配，同时把完整结构化快照保留在 Session 本地捕获中，供可选的 `ctx_execute_file` 分析使用。
-- click/type/fill/key 上的 `includeSnapshot=true` 会保存动作后的快照并返回其捕获路径，而不是把快照嵌入工具结果。
 - 授权后的 `before_agent_start` 不再返回 `systemPrompt`：Chrome 使用提示写入 `event.systemPromptOptions.sections.chrome_profile_bridge` 结构化分区，由组合包单一提示词出口 `extensions/system-prompt.ts` 统一组装，顺序由 `config/prompt/sections.jsonc` 控制；未授权时不贡献任何提示词内容。
+- 移除 `PI_CHROME_BRIDGE_HOST`/`PI_CHROME_BRIDGE_PORT` 环境变量，host/port 固定为 `127.0.0.1:17318`（Companion 扩展 `service_worker.js` 同样硬编码该地址，单侧可配只会断连且非回环 host 有安全风险）。
+- 移除 `MAX_TEXT_CHARS`/`truncateText` 的 30K 字符文本截断：chrome_* 工具返回完整文本，超长结果的展示截断由 pi-overflow 统一负责；`MAX_ELEMENTS` 等条数上限保留不变。
 
-同步上游时，请以导入 commit 为基准比较并选择性移植改动。不得整体覆盖这些本地生命周期、隐私和 Context Mode 集成。
+同步上游时，请以导入 commit 为基准比较并选择性移植改动。不得整体覆盖这些本地生命周期和组合包集成改动。

pi-chrome/extensions/chrome-profile-bridge/index.ts

@@ -4,11 +4,6 @@ import { existsSync, readFileSync, statSync } from "node:fs";
 import { mkdir, writeFile } from "node:fs/promises";
 import { createServer, type IncomingMessage, type Server, type ServerResponse } from "node:http";
 import { dirname, join, resolve } from "node:path";
-import {
- SessionSnapshotStore,
- type SnapshotCaptureMetadata,
- type SnapshotTarget,
-} from "../../src/session-snapshot-store.ts";
 
 /**
  * Existing-profile Chrome bridge for pi.
@@ -64,17 +59,11 @@ const PI_CHROME_GLOBAL_KEY = "__piChromeProfileBridgeLoaded__";
 // reload) so a /reload — which tears down and re-evaluates the module — does not silently drop
 // an active /chrome authorize grant.
 const PI_CHROME_AUTH_KEY = "__piChromeProfileBridgeAuth__";
-const DEFAULT_HOST = process.env.PI_CHROME_BRIDGE_HOST ?? "127.0.0.1";
+const DEFAULT_HOST = "127.0.0.1";
-const DEFAULT_PORT = Number(process.env.PI_CHROME_BRIDGE_PORT ?? "17318");
+const DEFAULT_PORT = 17318;
 const DEFAULT_TIMEOUT_MS = 30_000;
-const MAX_TEXT_CHARS = 30_000;
 const MAX_ELEMENTS = 80;
 
-function truncateText(text: string, maxChars = MAX_TEXT_CHARS): string {
-	if (text.length <= maxChars) return text;
-	return `${text.slice(0, maxChars)}\n\n[truncated ${text.length - maxChars} characters]`;
-}
-
 function safeJson(value: unknown): string {
 	return JSON.stringify(value, null, 2);
 }
@@ -95,13 +84,28 @@ function snapshotFromIncludedResult(raw: unknown): unknown {
 	return raw && typeof raw === "object" ? (raw as { snapshot?: unknown }).snapshot : undefined;
 }
 
-export function formatSnapshotCapture(metadata: SnapshotCaptureMetadata): string {
+/**
-	const kib = Math.max(1, Math.ceil(metadata.bytes / 1024));
+ * Small, session-file-safe metadata describing a returned snapshot. The full
-	return [
+ * snapshot goes to the model as text — details must stay small because Pi
-		`Chrome snapshot saved to ${metadata.path} (${kib} KiB, ${metadata.elementCount} elements${metadata.matchCount ? `, ${metadata.matchCount} matches` : ""}).`,
+ * persists them in the session file.
-		"Analyze it immediately with ctx_execute_file; FILE_CONTENT is JSON and the page data is under .snapshot.",
+ */
-		"This capture is replaced by the next snapshot for the same target and becomes stale after page interaction.",
+export function snapshotStats(snapshot: unknown): { elementCount: number; matchCount: number } {
-	].join("\n");
+	const object = snapshot && typeof snapshot === "object" && !Array.isArray(snapshot)
+		? snapshot as { elements?: unknown; matches?: unknown }
+		: {};
+	return {
+		elementCount: Array.isArray(object.elements) ? object.elements.length : 0,
+		matchCount: Array.isArray(object.matches) ? object.matches.length : 0,
+	};
+}
+
+/**
+ * Action summary followed by the inline snapshot JSON. Oversized results are
+ * bounded by pi-overflow and recoverable through overflow_query, so no local
+ * capture file is written.
+ */
+export function formatActionResultText(text: string, snapshot: unknown | undefined): string {
+	return snapshot === undefined ? text : `${text}\n\n${safeJson(snapshot)}`;
 }
 
 export function formatChromeFindMatches(snapshot: any): string {
@@ -142,7 +146,7 @@ function formatChromeInspect(inspect: any): string {
 		lines.push("\n## Ancestors");
 		for (const a of inspect.ancestors.slice(0, 6)) lines.push(`- ${a.uid} ${a.role || a.tag} ${compactLine(a.label || a.selector, 120)}`);
 	}
-	return truncateText(lines.join("\n"));
+	return lines.join("\n");
 }
 
 function extensionRoot(): string {
@@ -648,7 +652,6 @@ export default function (pi: ExtensionAPI): void {
 	let countdownInterval: NodeJS.Timeout | undefined;
 	// Remembered so bridge sends can tag tabs with this session's group even when ctx isn't handy.
 	let sessionCtx: ExtensionContext | undefined;
-	let snapshotStore: SessionSnapshotStore | undefined;
 
 	const clearAuthExpiryTimer = (): void => {
 		if (!authExpiryTimer) return;
@@ -722,7 +725,6 @@ export default function (pi: ExtensionAPI): void {
 		persistAuth();
 		// Revoking control ends pi-chrome's automation for this session; tidy up the target we own.
 		cleanupAutomationTargetBestEffort();
-		void snapshotStore?.clear().catch(() => undefined);
 	};
 
 	const authSummary = (): string => {
@@ -853,45 +855,22 @@ export default function (pi: ExtensionAPI): void {
 		return { ...params, foreground: !background } as T;
 	};
 
-	const snapshotTargetFrom = (params: Record<string, unknown>): SnapshotTarget => ({
+	const formatActionResult = (raw: unknown, params: Record<string, unknown>, text: string): ToolTextResult => {
-		targetId: typeof params.targetId === "string" ? params.targetId : undefined,
-		urlIncludes: typeof params.urlIncludes === "string" ? params.urlIncludes : undefined,
-		titleIncludes: typeof params.titleIncludes === "string" ? params.titleIncludes : undefined,
-	});
-
-	const saveSnapshotCapture = async (snapshot: unknown, params: Record<string, unknown>): Promise<SnapshotCaptureMetadata> => {
-		if (!snapshotStore) throw new Error("Chrome snapshot storage is unavailable before session_start.");
-		const stored = await snapshotStore.save(snapshot, snapshotTargetFrom(params));
-		return stored.metadata;
-	};
-
-	const invalidateSnapshotCapture = async (params?: Record<string, unknown>): Promise<void> => {
-		if (!snapshotStore) return;
-		if (!params) await snapshotStore.clear();
-		else await snapshotStore.invalidate(snapshotTargetFrom(params));
-	};
-
-	const formatActionWithCapture = async (raw: unknown, params: Record<string, unknown>, text: string): Promise<ToolTextResult> => {
 		const snapshot = snapshotFromIncludedResult(raw);
 		const result = params.includeSnapshot && raw && typeof raw === "object"
 			? (raw as { result?: unknown }).result
 			: raw;
 		if (!snapshot) {
-			await invalidateSnapshotCapture(params);
 			return { content: [{ type: "text", text }], details: { result: result as Json } };
 		}
-		const capture = await saveSnapshotCapture(snapshot, params);
 		return {
-			content: [{ type: "text", text: `${text}\n\n${formatSnapshotCapture(capture)}` }],
+			content: [{ type: "text", text: formatActionResultText(text, snapshot) }],
-			details: { result: result as Json, capture },
+			details: { result: result as Json, snapshot: snapshotStats(snapshot) },
 		};
 	};
 
 	pi.on("session_start", async (_event, ctx) => {
 		sessionCtx = ctx;
-		const sessionId = ctx.sessionManager.getSessionId?.() || "unknown-session";
-		snapshotStore = new SessionSnapshotStore(workspaceCwd(ctx), sessionId);
-		await snapshotStore.clear();
 		await bridge.start();
 		// Reestablish in-memory state after a /reload restored chromeAuthorizedUntil from globalThis.
 		if (chromeControlAuthorized()) {
@@ -909,10 +888,6 @@ export default function (pi: ExtensionAPI): void {
 	pi.on("session_shutdown", async (event) => {
 		clearAuthExpiryTimer();
 		clearCountdownInterval();
-		// Snapshot files are intentionally ephemeral. A reload must take a fresh observation too,
-		// because the browser page may have changed while the extension was reloading.
-		await snapshotStore?.clear().catch(() => undefined);
-		snapshotStore = undefined;
 		// Keep the owned automation target across /reload, but clean it up on real session end.
 		if (event?.reason !== "reload") cleanupAutomationTargetBestEffort();
 		bridge.stop();
@@ -939,12 +914,12 @@ Capability model (important):
 - Interactive controls (click/type/fill/key/hover/drag/scroll/tap) use Chrome's real input layer via chrome.debugger / CDP. Events satisfy normal user-activation gates.
 - Input bypasses page CSP because it is injected at browser input layer, not page JavaScript. Chrome may show the “Pi Chrome Connector started debugging this browser” banner while attached.
 - \`chrome_evaluate\` and \`chrome_snapshot\` run in MAIN world via **CDP \`Runtime.evaluate\`**, which is not subject to the page's Content-Security-Policy. They work even on strict-CSP pages (e.g. github.com, many bank/SaaS apps) that block \`'unsafe-eval'\`. \`chrome_navigate initScript\` likewise injects at document_start via CDP and bypasses CSP. \`chrome_screenshot\`, \`chrome_tab\`, and Chrome input also work under any CSP.
-- \`chrome_snapshot\` and snapshots requested with \`includeSnapshot=true\` are written to a private session-local JSON capture; raw page data is never returned inline. The tool result gives the path for immediate \`ctx_execute_file\` analysis.
+- \`chrome_snapshot\` and snapshots requested with \`includeSnapshot=true\` return the page snapshot as JSON directly in the tool result. When the result is large it carries a \`[pi-overflow id=…]\` marker — extract only the uids, text, or state you need with \`overflow_query\` (jsonPath such as \`elements[0].uid\` or \`summary.focused\`, or a \`search\`) instead of taking another snapshot.
 
 Usage rules:
 1. If a chrome_* tool says Chrome control is locked, ask the user to run \`/chrome authorize\` before retrying.
-2. Call \`chrome_snapshot\`, then immediately call \`ctx_execute_file\` on the returned path and print only the uids/state needed for the next action. Never use read/cat on the full capture.
+2. Call \`chrome_snapshot\` and read the returned JSON (or query a pi-overflow marker with \`overflow_query\`); keep only the uids/state needed for the next action.
-3. Prefer \`uid\` over \`selector\`. After any page interaction, treat the old capture as stale. Use \`includeSnapshot=true\` when a fresh post-action capture is needed.
+3. Prefer \`uid\` over \`selector\`. After any page interaction, earlier snapshots are stale — take a fresh one (or pass \`includeSnapshot=true\` on the action) instead of reusing old uids.
 4. If \`chrome_evaluate\` returns null when you expected a value, the expression evaluated to null/undefined in the page; surface the value via \`JSON.stringify\` to confirm.
 5. \`chrome_navigate\` supports an optional \`initScript\` that runs at document_start in MAIN world for the next navigation (good for seeding localStorage or stubbing Date.now).
 6. By default chrome_* tools run in the background without focusing Chrome; pass \`background=false\` or run /chrome background off when the user wants to watch Chrome work.
@@ -1337,7 +1312,6 @@ Usage rules:
 				const text = tabs.map((tab) => `${tab.id}\t${tab.active ? "*" : " "}\t${tab.group?.title ? `[${tab.group.title}] ` : ""}${tab.title || "(untitled)"}\t${tab.url}`).join("\n") || "No tabs.";
 				return { content: [{ type: "text", text }], details: { tabs } };
 			}
-			if (params.action !== "version") await invalidateSnapshotCapture();
 			return { content: [{ type: "text", text: safeJson(result) }], details: { result: result as Json } };
 		},
 	});
@@ -1346,8 +1320,8 @@ Usage rules:
 		name: "chrome_snapshot",
 		label: "Chrome Snapshot",
 		description:
-			"Capture the current Chrome page into a private, session-local JSON file instead of placing the raw DOM snapshot in model context. The result returns a path for immediate ctx_execute_file analysis. Each target keeps only its latest capture; page interaction makes it stale. Use mode/query/nearUid to control what Chrome captures. Runs in the background by default; pass background=false so the user can watch.",
+			"Capture the current Chrome page and return the snapshot as JSON directly in the result. When it is large the result carries a [pi-overflow id=…] marker — extract only the uids, text, or state you need with overflow_query (jsonPath or search). Page interaction makes a snapshot stale; take a fresh one after navigating or acting. Use mode/query/nearUid to control what Chrome captures. Runs in the background by default; pass background=false so the user can watch.",
-		promptSnippet: "Save the current Chrome page to a session-local JSON capture, then analyze that path with ctx_execute_file for only the required uids and state.",
+		promptSnippet: "Capture the current Chrome page and return its JSON snapshot; query large results via overflow_query.",
 		parameters: Type.Object({
 			targetId: Type.Optional(Type.String()),
 			urlIncludes: Type.Optional(Type.String()),
@@ -1372,8 +1346,7 @@ Usage rules:
 				DEFAULT_TIMEOUT_MS,
 				signal,
 			);
-			const capture = await saveSnapshotCapture(snapshot, params);
+			return { content: [{ type: "text", text: safeJson(snapshot) }], details: { snapshot: snapshotStats(snapshot) } };
-			return { content: [{ type: "text", text: formatSnapshotCapture(capture) }], details: { capture } };
 		},
 	});
 
@@ -1381,8 +1354,8 @@ Usage rules:
 		name: "chrome_find",
 		label: "Chrome Find",
 		description:
-			"Find elements, regions, or text on the live Chrome page. Returns only compact ranked matches with stable uids and coordinates; the full structured snapshot is saved to the same private session-local file used by chrome_snapshot for optional ctx_execute_file analysis.",
+			"Find elements, regions, or text on the live Chrome page. Returns only compact ranked matches with stable uids and coordinates.",
-		promptSnippet: "Find live Chrome controls/text/regions and return compact stable-uid matches while keeping the full snapshot out of model context.",
+		promptSnippet: "Find live Chrome controls/text/regions and return compact stable-uid matches.",
 		parameters: Type.Object({
 			query: Type.String({ description: "What to find, e.g. 'merge button', 'email error', 'approve PR', 'search box'." }),
 			mode: Type.Optional(StringEnum(snapshotModeValues)),
@@ -1403,10 +1376,9 @@ Usage rules:
 				DEFAULT_TIMEOUT_MS,
 				signal,
 			);
-			const capture = await saveSnapshotCapture(snapshot, params);
 			return {
-				content: [{ type: "text", text: `${formatChromeFindMatches(snapshot)}\n\n${formatSnapshotCapture(capture)}` }],
+				content: [{ type: "text", text: formatChromeFindMatches(snapshot) }],
-				details: { capture },
+				details: { snapshot: snapshotStats(snapshot) },
 			};
 		},
 	});
@@ -1452,9 +1424,8 @@ Usage rules:
 					DEFAULT_TIMEOUT_MS,
 					signal,
 				);
-				const capture = await saveSnapshotCapture(snapshot, params);
+				const text = `chrome_inspect fallback: loaded Chrome extension does not yet support page.inspect; reload it at chrome://extensions for deep inspect.\n\n${safeJson(snapshot)}`;
-				const text = `chrome_inspect fallback: loaded Chrome extension does not yet support page.inspect; reload it at chrome://extensions for deep inspect.\n\n${formatSnapshotCapture(capture)}`;
+				return { content: [{ type: "text", text }], details: { snapshot: snapshotStats(snapshot), fallback: "page.snapshot" } };
-				return { content: [{ type: "text", text }], details: { capture, fallback: "page.snapshot" } };
 			}
 		},
 	});
@@ -1481,7 +1452,6 @@ Usage rules:
 		}),
 		async execute(_id, params, signal): Promise<ToolTextResult> {
 			const result = await authorizedBridgeSend("page.navigate", withBackground(params), (params.timeoutMs ?? 15_000) + 2_000, signal);
-			await invalidateSnapshotCapture(params);
 			return { content: [{ type: "text", text: `Navigated to ${params.url}${params.initScript ? " (with initScript)" : ""}` }], details: { result: result as Json } };
 		},
 	});
@@ -1506,13 +1476,12 @@ Usage rules:
 		}),
 		async execute(_id, params, signal): Promise<ToolTextResult> {
 			const value = await authorizedBridgeSend("page.evaluate", withBackground(params), DEFAULT_TIMEOUT_MS, signal);
-			await invalidateSnapshotCapture(params);
 			const text = value === undefined
 				? "undefined"
 				: typeof value === "string"
 					? value
 					: safeJson(value) ?? "undefined";
-			return { content: [{ type: "text", text: truncateText(text) }], details: { value: value as Json } };
+			return { content: [{ type: "text", text }], details: { value: value as Json } };
 		},
 	});
 
@@ -1520,7 +1489,7 @@ Usage rules:
 		name: "chrome_click",
 		label: "Chrome Click",
 		description:
-			"Click a snapshot uid, CSS selector, or viewport coordinate using Chrome's real input layer. The previous capture becomes stale. Pass includeSnapshot=true to save a fresh post-click snapshot to the session-local file and return its ctx_execute_file path, never the raw snapshot.",
+			"Click a snapshot uid, CSS selector, or viewport coordinate using Chrome's real input layer. Earlier snapshots become stale after the click. Pass includeSnapshot=true to get a fresh post-click snapshot JSON inline.",
 		promptSnippet: "Click page elements in Chrome by snapshot uid, selector, or viewport coordinate.",
 		parameters: Type.Object({
 			uid: Type.Optional(Type.String({ description: "Stable element uid from chrome_snapshot. Prefer uid over selector after taking a snapshot." })),
@@ -1528,7 +1497,7 @@ Usage rules:
 			x: Type.Optional(Type.Number({ description: "Viewport x coordinate if uid/selector is omitted." })),
 			y: Type.Optional(Type.Number({ description: "Viewport y coordinate if uid/selector is omitted." })),
 			domFallback: Type.Optional(Type.Boolean({ description: "If true (default), fall back to DOM-dispatched click if Chrome's CDP input path is blocked by another extension overlay or debugger failure." })),
-			includeSnapshot: Type.Optional(Type.Boolean({ description: "If true, save a fresh post-click snapshot to the session-local capture file and return its path." })),
+			includeSnapshot: Type.Optional(Type.Boolean({ description: "If true, return a fresh post-click snapshot JSON inline after the action summary." })),
 			maxElements: Type.Optional(Type.Number({ default: MAX_ELEMENTS, description: "Max elements in the included snapshot." })),
 			targetId: Type.Optional(Type.String()),
 			urlIncludes: Type.Optional(Type.String()),
@@ -1545,7 +1514,7 @@ Usage rules:
 			const summary = summarizeActionResult(result);
 			const target = params.uid ?? params.selector ?? `${params.x},${params.y}`;
 			const text = summary ? `Clicked ${target} — ${summary}` : `Clicked ${target}`;
-			return formatActionWithCapture(raw, params, text);
+			return formatActionResult(raw, params, text);
 		},
 	});
 
@@ -1553,13 +1522,13 @@ Usage rules:
 		name: "chrome_type",
 		label: "Chrome Type",
 		description:
-			"Focus an optional snapshot uid or CSS selector, then type text using Chrome's real keyboard input. The previous capture becomes stale. Pass includeSnapshot=true to save a fresh post-type snapshot and return its ctx_execute_file path.",
+			"Focus an optional snapshot uid or CSS selector, then type text using Chrome's real keyboard input. Earlier snapshots become stale after typing. Pass includeSnapshot=true to get a fresh post-type snapshot JSON inline.",
 		promptSnippet: "Type text into Chrome, optionally focusing a snapshot uid or selector first.",
 		parameters: Type.Object({
 			text: Type.String(),
 			uid: Type.Optional(Type.String({ description: "Stable element uid from chrome_snapshot." })),
 			selector: Type.Optional(Type.String({ description: "CSS selector to focus before typing." })),
-			includeSnapshot: Type.Optional(Type.Boolean({ description: "If true, save a fresh post-type snapshot to the session-local capture file and return its path." })),
+			includeSnapshot: Type.Optional(Type.Boolean({ description: "If true, return a fresh post-type snapshot JSON inline after the action summary." })),
 			maxElements: Type.Optional(Type.Number({ default: MAX_ELEMENTS, description: "Max elements in the included snapshot." })),
 			pressEnter: Type.Optional(Type.Boolean()),
 			targetId: Type.Optional(Type.String()),
@@ -1578,7 +1547,7 @@ Usage rules:
 			const into = params.uid || params.selector ? ` into ${params.uid ?? params.selector}` : "";
 			const base = `Typed ${params.text.length} character(s)${into}.`;
 			const text = summary ? `${base} (${summary})` : base;
-			return formatActionWithCapture(raw, params, text);
+			return formatActionResult(raw, params, text);
 		},
 	});
 
@@ -1586,7 +1555,7 @@ Usage rules:
 		name: "chrome_fill",
 		label: "Chrome Fill",
 		description:
-			"Set the full value of a text input, textarea, or contenteditable element using Chrome input. The previous capture becomes stale. Pass includeSnapshot=true to save a fresh post-fill snapshot and return its ctx_execute_file path.",
+			"Set the full value of a text input, textarea, or contenteditable element using Chrome input. Earlier snapshots become stale after filling. Pass includeSnapshot=true to get a fresh post-fill snapshot JSON inline.",
 		promptSnippet: "Fill a Chrome form field by snapshot uid or selector, optionally returning a fresh snapshot.",
 		parameters: Type.Object({
 			text: Type.String(),
@@ -1594,7 +1563,7 @@ Usage rules:
 			selector: Type.Optional(Type.String({ description: "CSS selector to fill if uid is omitted." })),
 			submit: Type.Optional(Type.Boolean({ description: "If true, press Enter after filling." })),
 			domFallback: Type.Optional(Type.Boolean({ description: "If true (default), fall back to DOM value-setting if Chrome's CDP input path is blocked by another extension overlay or debugger failure." })),
-			includeSnapshot: Type.Optional(Type.Boolean({ description: "If true, save a fresh post-fill snapshot to the session-local capture file and return its path." })),
+			includeSnapshot: Type.Optional(Type.Boolean({ description: "If true, return a fresh post-fill snapshot JSON inline after the action summary." })),
 			maxElements: Type.Optional(Type.Number({ default: MAX_ELEMENTS, description: "Max elements in the included snapshot." })),
 			targetId: Type.Optional(Type.String()),
 			urlIncludes: Type.Optional(Type.String()),
@@ -1612,7 +1581,7 @@ Usage rules:
 			const into = params.uid || params.selector ? ` into ${params.uid ?? params.selector}` : "";
 			const base = `Filled ${params.text.length} character(s)${into}.`;
 			const text = summary ? `${base} (${summary})` : base;
-			return formatActionWithCapture(raw, params, text);
+			return formatActionResult(raw, params, text);
 		},
 	});
 
@@ -1620,7 +1589,7 @@ Usage rules:
 		name: "chrome_key",
 		label: "Chrome Key",
 		description:
-			"Send a keyboard key to an existing Chrome tab. The previous capture becomes stale. Pass includeSnapshot=true to save a fresh post-key snapshot and return its ctx_execute_file path. Runs in the background by default; pass background=false so the user can watch.",
+			"Send a keyboard key to an existing Chrome tab. Earlier snapshots become stale after the key press. Pass includeSnapshot=true to get a fresh post-key snapshot JSON inline. Runs in the background by default; pass background=false so the user can watch.",
 		promptSnippet: "Press keys in Chrome through the companion extension.",
 		parameters: Type.Object({
 			key: Type.String(),
@@ -1630,7 +1599,7 @@ Usage rules:
 				altKey: Type.Optional(Type.Boolean()),
 				metaKey: Type.Optional(Type.Boolean()),
 			}, { description: "Modifier keys to hold while pressing the key (chord)." })),
-			includeSnapshot: Type.Optional(Type.Boolean({ description: "If true, save a fresh post-key snapshot to the session-local capture file and return its path." })),
+			includeSnapshot: Type.Optional(Type.Boolean({ description: "If true, return a fresh post-key snapshot JSON inline after the action summary." })),
 			maxElements: Type.Optional(Type.Number({ default: MAX_ELEMENTS, description: "Max elements in the included snapshot." })),
 			targetId: Type.Optional(Type.String()),
 			urlIncludes: Type.Optional(Type.String()),
@@ -1647,7 +1616,7 @@ Usage rules:
 			const summary = summarizeActionResult(result);
 			const base = `Pressed ${params.key}.`;
 			const text = summary ? `${base} (${summary})` : base;
-			return formatActionWithCapture(raw, params, text);
+			return formatActionResult(raw, params, text);
 		},
 	});
 
@@ -1669,7 +1638,6 @@ Usage rules:
 		}),
 		async execute(_id, params, signal): Promise<ToolTextResult> {
 			const result = await authorizedBridgeSend("page.waitFor", params, (params.timeoutMs ?? 10_000) + 2_000, signal);
-			await invalidateSnapshotCapture(params);
 			return { content: [{ type: "text", text: `Observed ${params.kind}: ${params.value}` }], details: { result: result as Json } };
 		},
 	});
@@ -1691,7 +1659,7 @@ Usage rules:
 		}),
 		async execute(_id, params, signal): Promise<ToolTextResult> {
 			const result = await authorizedBridgeSend("page.console.list", withBackground(params), DEFAULT_TIMEOUT_MS, signal);
-			return { content: [{ type: "text", text: truncateText(safeJson(result)) }], details: { result: result as Json } };
+			return { content: [{ type: "text", text: safeJson(result) }], details: { result: result as Json } };
 		},
 	});
 
@@ -1713,7 +1681,7 @@ Usage rules:
 		}),
 		async execute(_id, params, signal): Promise<ToolTextResult> {
 			const result = await authorizedBridgeSend("page.network.list", withBackground(params), DEFAULT_TIMEOUT_MS, signal);
-			return { content: [{ type: "text", text: truncateText(safeJson(result)) }], details: { result: result as Json } };
+			return { content: [{ type: "text", text: safeJson(result) }], details: { result: result as Json } };
 		},
 	});
 
@@ -1733,7 +1701,7 @@ Usage rules:
 		}),
 		async execute(_id, params, signal): Promise<ToolTextResult> {
 			const result = await authorizedBridgeSend("page.network.get", withBackground(params), DEFAULT_TIMEOUT_MS, signal);
-			return { content: [{ type: "text", text: truncateText(safeJson(result)) }], details: { result: result as Json } };
+			return { content: [{ type: "text", text: safeJson(result) }], details: { result: result as Json } };
 		},
 	});
 
@@ -1813,7 +1781,6 @@ Usage rules:
 		}),
 		async execute(_id, params, signal): Promise<ToolTextResult> {
 			const result = await authorizedBridgeSend("page.hover", withBackground(params), DEFAULT_TIMEOUT_MS, signal);
-			await invalidateSnapshotCapture(params);
 			return { content: [{ type: "text", text: `Hovered ${params.uid ?? params.selector ?? `${params.x},${params.y}`}` }], details: { result: result as Json } };
 		},
 	});
@@ -1840,7 +1807,6 @@ Usage rules:
 		}),
 		async execute(_id, params, signal): Promise<ToolTextResult> {
 			const result = await authorizedBridgeSend("page.drag", withBackground(params), DEFAULT_TIMEOUT_MS, signal);
-			await invalidateSnapshotCapture(params);
 			return { content: [{ type: "text", text: `Dragged from ${params.fromUid ?? params.fromSelector} to ${params.toUid ?? params.toSelector}` }], details: { result: result as Json } };
 		},
 	});
@@ -1863,7 +1829,6 @@ Usage rules:
 		}),
 		async execute(_id, params, signal): Promise<ToolTextResult> {
 			const result = await authorizedBridgeSend("page.tap", withBackground(params), DEFAULT_TIMEOUT_MS, signal);
-			await invalidateSnapshotCapture(params);
 			const target = params.uid ?? params.selector ?? `${params.x},${params.y}`;
 			return { content: [{ type: "text", text: `Tapped ${target} (touch)` }], details: { result: result as Json } };
 		},
@@ -1887,7 +1852,6 @@ Usage rules:
 		}),
 		async execute(_id, params, signal): Promise<ToolTextResult> {
 			const result = await authorizedBridgeSend("page.scroll", withBackground(params), DEFAULT_TIMEOUT_MS, signal);
-			await invalidateSnapshotCapture(params);
 			return { content: [{ type: "text", text: `Scrolled dy=${params.deltaY ?? 0} dx=${params.deltaX ?? 0}` }], details: { result: result as Json } };
 		},
 	});
@@ -1910,7 +1874,6 @@ Usage rules:
 			const cwd = workspaceCwd(ctx);
 			const paths = params.paths.map((p) => resolve(cwd, p));
 			const result = await authorizedBridgeSend("page.upload", withBackground({ ...params, paths }), DEFAULT_TIMEOUT_MS, signal);
-			await invalidateSnapshotCapture(params);
 			return { content: [{ type: "text", text: `Uploaded ${paths.length} file(s) to ${params.uid ?? params.selector}` }], details: { result: result as Json } };
 		},
 	});

pi-chrome/package.json

@@ -2,7 +2,7 @@
 	"name": "pi-chrome",
 	"version": "0.15.46",
 	"scripts": {
-		"test": "node test-suite/unit/csp-eval.test.mjs && node test-suite/unit/automation-target.test.mjs && jiti test/session-snapshot-store.test.ts",
+		"test": "node test-suite/unit/csp-eval.test.mjs && node test-suite/unit/automation-target.test.mjs && jiti test/snapshot-result.test.ts",
 		"version": "node scripts/sync-manifest-version.js",
 		"prepublishOnly": "node scripts/sync-manifest-version.js"
 	},
@@ -52,7 +52,6 @@
 	"type": "commonjs",
 	"files": [
 		"extensions",
-		"src",
 		"test",
 		"docs",
 		"test-suite",

pi-chrome/src/session-snapshot-store.ts

@@ -1,140 +0,0 @@
-import { createHash, randomUUID } from "node:crypto";
-import { chmod, mkdir, rename, rm, writeFile } from "node:fs/promises";
-import { join, relative, resolve, sep } from "node:path";
-
-type SnapshotObject = Record<string, unknown>;
-
-export type SnapshotTarget = {
-	targetId?: string;
-	urlIncludes?: string;
-	titleIncludes?: string;
-};
-
-export type SnapshotCaptureMetadata = {
-	version: 1;
-	captureId: string;
-	capturedAt: string;
-	path: string;
-	bytes: number;
-	targetKey: string;
-	targetId?: string | number;
-	mode?: string;
-	elementCount: number;
-	matchCount: number;
-	staleAfterInteraction: true;
-};
-
-export type StoredSnapshotCapture = {
-	metadata: SnapshotCaptureMetadata;
-	absolutePath: string;
-};
-
-export type SnapshotCaptureFile = {
-	version: 1;
-	captureId: string;
-	capturedAt: string;
-	targetKey: string;
-	snapshot: unknown;
-};
-
-function hashSegment(value: string): string {
-	return createHash("sha256").update(value).digest("hex").slice(0, 20);
-}
-
-function normalizeRelativePath(path: string): string {
-	return path.split(sep).join("/");
-}
-
-function snapshotObject(snapshot: unknown): SnapshotObject | undefined {
-	return snapshot && typeof snapshot === "object" && !Array.isArray(snapshot)
-		? snapshot as SnapshotObject
-		: undefined;
-}
-
-export function snapshotTargetKey(target: SnapshotTarget = {}): string {
-	if (target.targetId) return `target:${target.targetId}`;
-	if (target.urlIncludes) return `url:${target.urlIncludes}`;
-	if (target.titleIncludes) return `title:${target.titleIncludes}`;
-	return "default";
-}
-
-export class SessionSnapshotStore {
-	readonly workspaceRoot: string;
-	readonly sessionRoot: string;
-	readonly sessionHash: string;
-
-	constructor(workspaceRoot: string, sessionId: string) {
-		this.workspaceRoot = resolve(workspaceRoot);
-		this.sessionHash = hashSegment(sessionId || "unknown-session");
-		this.sessionRoot = join(this.workspaceRoot, ".pi", "chrome-context", this.sessionHash);
-	}
-
-	async save(snapshot: unknown, target: SnapshotTarget = {}): Promise<StoredSnapshotCapture> {
-		const targetKey = snapshotTargetKey(target);
-		const targetDirectory = join(this.sessionRoot, hashSegment(targetKey));
-		const absolutePath = join(targetDirectory, "snapshot.json");
-		const temporaryPath = join(targetDirectory, `.snapshot.${process.pid}.${randomUUID()}.tmp`);
-		const captureId = randomUUID();
-		const capturedAt = new Date().toISOString();
-		const capture: SnapshotCaptureFile = {
-			version: 1,
-			captureId,
-			capturedAt,
-			targetKey,
-			snapshot,
-		};
-		const serialized = `${JSON.stringify(capture, null, 2)}\n`;
-
-		await mkdir(targetDirectory, { recursive: true, mode: 0o700 });
-		await chmod(this.sessionRoot, 0o700).catch(() => undefined);
-		await chmod(targetDirectory, 0o700).catch(() => undefined);
-		try {
-			await writeFile(temporaryPath, serialized, { encoding: "utf8", mode: 0o600, flag: "wx" });
-			await chmod(temporaryPath, 0o600).catch(() => undefined);
-			await rename(temporaryPath, absolutePath);
-			await chmod(absolutePath, 0o600).catch(() => undefined);
-		} catch (error) {
-			await rm(temporaryPath, { force: true }).catch(() => undefined);
-			throw error;
-		}
-
-		const object = snapshotObject(snapshot);
-		const elements = Array.isArray(object?.elements) ? object.elements : [];
-		const matches = Array.isArray(object?.matches) ? object.matches : [];
-		const targetId = typeof object?.targetId === "string" || typeof object?.targetId === "number"
-			? object.targetId
-			: undefined;
-		const mode = typeof object?.mode === "string" ? object.mode : undefined;
-		const path = normalizeRelativePath(relative(this.workspaceRoot, absolutePath));
-
-		return {
-			absolutePath,
-			metadata: {
-				version: 1,
-				captureId,
-				capturedAt,
-				path,
-				bytes: Buffer.byteLength(serialized),
-				targetKey,
-				targetId,
-				mode,
-				elementCount: elements.length,
-				matchCount: matches.length,
-				staleAfterInteraction: true,
-			},
-		};
-	}
-
-	async invalidate(target?: SnapshotTarget): Promise<void> {
-		if (!target) {
-			await this.clear();
-			return;
-		}
-		const targetDirectory = join(this.sessionRoot, hashSegment(snapshotTargetKey(target)));
-		await rm(targetDirectory, { recursive: true, force: true });
-	}
-
-	async clear(): Promise<void> {
-		await rm(this.sessionRoot, { recursive: true, force: true });
-	}
-}

pi-chrome/test/session-snapshot-store.test.ts

@@ -1,100 +0,0 @@
-import assert from "node:assert/strict";
-import { mkdtemp, readFile, stat } from "node:fs/promises";
-import { tmpdir } from "node:os";
-import { join } from "node:path";
-import test from "node:test";
-import { SessionSnapshotStore, snapshotTargetKey } from "../src/session-snapshot-store.ts";
-import { CHROME_TOOL_NAMES, formatChromeFindMatches, formatSnapshotCapture } from "../extensions/chrome-profile-bridge/index.ts";
-
-test("snapshotTargetKey prefers explicit target selectors", () => {
-	assert.equal(snapshotTargetKey(), "default");
-	assert.equal(snapshotTargetKey({ titleIncludes: "Inbox" }), "title:Inbox");
-	assert.equal(snapshotTargetKey({ urlIncludes: "example.com" }), "url:example.com");
-	assert.equal(snapshotTargetKey({ targetId: "42", urlIncludes: "ignored" }), "target:42");
-});
-
-test("SessionSnapshotStore atomically keeps only the latest capture per target", async () => {
-	const workspace = await mkdtemp(join(tmpdir(), "pi-chrome-snapshot-"));
-	const store = new SessionSnapshotStore(workspace, "session/private/id");
-	const first = await store.save({ mode: "interactive", targetId: 42, elements: [{ uid: "el-1" }], matches: [] });
-	const second = await store.save({ mode: "changes", targetId: 42, elements: [{ uid: "el-2" }], matches: [{ uid: "el-2" }] });
-
-	assert.equal(first.metadata.path, second.metadata.path);
-	assert.notEqual(first.metadata.captureId, second.metadata.captureId);
-	assert.equal(second.metadata.elementCount, 1);
-	assert.equal(second.metadata.matchCount, 1);
-	assert.equal(second.metadata.targetId, 42);
-	assert.ok(!second.metadata.path.includes("private"));
-
-	const stored = JSON.parse(await readFile(second.absolutePath, "utf8"));
-	assert.equal(stored.version, 1);
-	assert.equal(stored.captureId, second.metadata.captureId);
-	assert.equal(stored.snapshot.elements[0].uid, "el-2");
-
-	if (process.platform !== "win32") {
-		assert.equal((await stat(store.sessionRoot)).mode & 0o777, 0o700);
-		assert.equal((await stat(second.absolutePath)).mode & 0o777, 0o600);
-	}
-
-	await store.clear();
-	await assert.rejects(stat(second.absolutePath), { code: "ENOENT" });
-});
-
-test("target invalidation and session cleanup do not affect another session", async () => {
-	const workspace = await mkdtemp(join(tmpdir(), "pi-chrome-isolation-"));
-	const firstSession = new SessionSnapshotStore(workspace, "session-a");
-	const secondSession = new SessionSnapshotStore(workspace, "session-b");
-	const firstTarget = { targetId: "101" };
-	const secondTarget = { targetId: "202" };
-	const first = await firstSession.save({ elements: [] }, firstTarget);
-	const otherTarget = await firstSession.save({ elements: [] }, secondTarget);
-	const otherSession = await secondSession.save({ elements: [] }, firstTarget);
-
-	await firstSession.invalidate(firstTarget);
-	await assert.rejects(stat(first.absolutePath), { code: "ENOENT" });
-	await stat(otherTarget.absolutePath);
-	await stat(otherSession.absolutePath);
-
-	await firstSession.clear();
-	await assert.rejects(stat(otherTarget.absolutePath), { code: "ENOENT" });
-	await stat(otherSession.absolutePath);
-	await secondSession.clear();
-});
-
-test("snapshot result formatting stays bounded and points to Context Mode", () => {
-	const text = formatSnapshotCapture({
-		version: 1,
-		captureId: "capture-1",
-		capturedAt: "2026-08-28T00:00:00.000Z",
-		path: ".pi/chrome-context/session/target/snapshot.json",
-		bytes: 80_000,
-		targetKey: "target:42",
-		targetId: 42,
-		mode: "interactive",
-		elementCount: 80,
-		matchCount: 3,
-		staleAfterInteraction: true,
-	});
-
-	assert.match(text, /ctx_execute_file/);
-	assert.match(text, /\.pi\/chrome-context\/session\/target\/snapshot\.json/);
-	assert.doesNotMatch(text, /raw-secret-page-content/);
-});
-
-test("chrome_find formatting is capped and authorization tracks all 21 tools", () => {
-	const matches = Array.from({ length: 25 }, (_, index) => ({
-		kind: "element",
-		uid: `el-${index}`,
-		role: "button",
-		label: `Button ${index}`,
-		rect: { x: index, y: index, width: 10, height: 10 },
-	}));
-	const text = formatChromeFindMatches({ query: "button", matches });
-
-	assert.match(text, /el-0/);
-	assert.match(text, /el-19/);
-	assert.doesNotMatch(text, /el-20/);
-	assert.equal(CHROME_TOOL_NAMES.length, 21);
-	assert.ok(CHROME_TOOL_NAMES.includes("chrome_find"));
-	assert.ok(CHROME_TOOL_NAMES.includes("chrome_inspect"));
-});

pi-chrome/test/snapshot-result.test.ts

@@ -0,0 +1,48 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+import {
+	CHROME_TOOL_NAMES,
+	formatActionResultText,
+	formatChromeFindMatches,
+	snapshotStats,
+} from "../extensions/chrome-profile-bridge/index.ts";
+
+test("formatActionResultText appends the inline snapshot JSON", () => {
+	const snapshot = { elements: [{ uid: "el-1", role: "button", label: "Save" }], matches: [] };
+	const text = formatActionResultText("Clicked el-1", snapshot);
+	const [summary, json] = text.split("\n\n");
+	assert.equal(summary, "Clicked el-1");
+	assert.deepEqual(JSON.parse(json), snapshot);
+});
+
+test("formatActionResultText returns the summary alone without a snapshot", () => {
+	assert.equal(formatActionResultText("Clicked el-1", undefined), "Clicked el-1");
+});
+
+test("snapshotStats stays small: only element and match counts", () => {
+	assert.deepEqual(
+		snapshotStats({ elements: [{ uid: "a" }, { uid: "b" }], matches: [{ uid: "a" }] }),
+		{ elementCount: 2, matchCount: 1 },
+	);
+	assert.deepEqual(snapshotStats({ elements: "nope" }), { elementCount: 0, matchCount: 0 });
+	assert.deepEqual(snapshotStats(undefined), { elementCount: 0, matchCount: 0 });
+	assert.deepEqual(snapshotStats([1, 2, 3]), { elementCount: 0, matchCount: 0 });
+});
+
+test("chrome_find formatting is capped and authorization tracks all 21 tools", () => {
+	const matches = Array.from({ length: 25 }, (_, index) => ({
+		kind: "element",
+		uid: `el-${index}`,
+		role: "button",
+		label: `Button ${index}`,
+		rect: { x: index, y: index, width: 10, height: 10 },
+	}));
+	const text = formatChromeFindMatches({ query: "button", matches });
+
+	assert.match(text, /el-0/);
+	assert.match(text, /el-19/);
+	assert.doesNotMatch(text, /el-20/);
+	assert.equal(CHROME_TOOL_NAMES.length, 21);
+	assert.ok(CHROME_TOOL_NAMES.includes("chrome_find"));
+	assert.ok(CHROME_TOOL_NAMES.includes("chrome_inspect"));
+});

pi-compaction/AGENTS.md

@@ -1,19 +1,23 @@
 # my-pi 维护约定（pi-compaction）
 
-- 扩展入口是 `pi-compaction/index.ts`，增量索引、投影、后台摘要、服务与 Codex native 实现在 `src/`，测试位于 `test/`，双上游来源记录是 `UPSTREAM.md`。
+- 扩展入口是 `pi-compaction/index.ts`，遮蔽规划、投影索引、遮蔽台账、服务与 Codex native 实现在 `src/`，测试位于 `test/`，双上游来源记录是 `UPSTREAM.md`。
 - 根包必须通过 `file:./pi-compaction` 提供依赖并直接加载源码；不得重新声明或加载 `pi-condense`、npm `@ogulcancelik/pi-codex-compaction` 或 npm `pi-context-view`。
-- `pi-compaction` 只支持自己的 version 1 记录和 `config/pi-compaction.json` 配置，不读取或迁移旧 `context-prune-*`、`contextPrune` 或 `openai-codex-native-compaction` 状态；切换后应使用新 Session。配置不写回 `settings.json`，默认 soft/hard/target 比例为 72%/90%/55%。
+- `pi-compaction` 只支持自己的 version 1 记录和 `config/pi-compaction.json` 配置，不读取或迁移旧 `context-prune-*`、`contextPrune`、`openai-codex-native-compaction` 或旧 `segment-summary` 状态；切换后应使用新 Session。配置不写回 `settings.json`。
-- Context Authority 是 compaction 唯一生产投影者。普通 Provider、Codex native、`/context`、`/compaction-status` 和 footer 必须消费同一个不可变 projection snapshot；`pi-minimal-footer` 优先显示同一 effective usage，snapshot 尚未建立时才回退宿主原始 usage；background worker 不得在 Provider 请求中被 await。
+- Context Authority 是 compaction 唯一生产投影者。普通 Provider、Codex native、`/context`、`/compaction-status` 和 footer 必须消费同一个不可变 projection snapshot；`pi-minimal-footer` 优先显示同一 effective usage，snapshot 尚未建立时才回退宿主原始 usage。
-- `pi-compaction` 负责已经进入 Session 的历史投影，与 Context Mode 的输入隔离职责互补。首次 Session hydration 允许一次完整扫描；后续普通 `context` hot path 必须复用不可变 snapshot 或只处理 delta，不得调用 summarizer、写 Session 或重新构建完整 branch。
+- `pi-compaction` 负责已经进入 Session 的历史投影，与 pi-overflow 的新结果入口截断职责互补。首次 Session hydration 允许一次完整扫描；后续普通 `context` hot path 必须复用不可变 snapshot 或只处理 delta，不得调用 summarizer、写 Session 或重新构建完整 branch。
-- 已闭合工具结果先同步生成保留工具名、参数、状态、首尾证据和 source key 的有界确定性 stub；大于 `semantic.minInputChars` 的结果再进入单 Session 有界后台语义 worker。后台结果只在安全事件边界提交，陈旧 hash 必须丢弃，Provider 请求不得等待后台摘要。
+- L2 遮蔽只用 Pi 原生 `context_edit`（`{ type: "context_edit", targetId, replacement }`）处理保护窗口外的旧 toolResult：成批提交、受 trigger/target 回差与 `minClearTokens` 约束，不逐条即时替换；不为遮蔽写 `segment-summary` 等 custom entry，原始条目、UI 与导出保持不变。
-- Segment custom entry 只能保存摘要、hash 和 source reference，不得复制完整原始工具输出。`context_tree_query` 恢复结果必须有界，native checkpoint 的 portable summary 也必须有界。
+- 不得改写最近 `keepRecentTurns` 轮；不得遮蔽 user/assistant 消息、`excludeTools` 中的工具、`SKILL.md` 的 `read` 结果、已遮蔽或已省略的条目、以及小于 `minEntryTokens` 的结果。
-- effective token 默认使用 Provider usage-backed baseline 加 projected delta；无可靠 baseline 时使用带上下界的估算。Pi Core raw threshold 只负责唤醒，只有 effective lower bound 达到 `hardRatio` 才执行 threshold compaction；manual 与 overflow 不得被阻断。
+- 遮蔽占位符必须确定性、单行、以 `MASK_PREFIX`（`"[pi-compaction] Earlier "`）开头，并在文本中指向 `context_tree_query` 恢复；`context_tree_query` 与 native checkpoint 的 portable summary 都必须有界。被遮蔽的 toolResult 若带 `details.overflow.id`（pi-overflow 已存原文），占位符末尾追加 `overflow_query` 提示——只读该公开字段，不得 import pi-overflow 内部实现。
+- L3 整段摘要只能从空闲的 `agent_settled` 通过 `ctx.compact` 触发（`ctx.compact` 会中止当前运行，非空闲时调用会破坏运行）；同一 leaf 不重复触发。阈值以绝对 token 计并受窗口比例约束（`effectiveThreshold = min(tokens, floor(ratio × contextWindow))`）。
+- `context` 钩子不得改写消息——非 checkpoint 路径只更新 snapshot 并返回 undefined；唯一例外是 native checkpoint 路径：checkpoint 之后的条目在投影时必须应用各 target 最新的 `context_edit`（null 省略、字符串规范成 text block），因为 `sessionEntryToContextMessages` 不处理 edits。
+- `session_before_compact` 只负责生成 portable/native 摘要，不得按自身阈值取消 Pi 的压缩（Pi 的 `contextTokens > window - reserveTokens` 是最终安全网）。
+- effective token 默认使用 Provider usage-backed baseline 加 projected delta；无可靠 baseline 时使用带上下界的估算。遮蔽台账在 `session_start` / `session_tree` / `session_compact` 从当前 branch 的 `context_edit` 条目重建，指标暴露 `maskedResults` 与 `reclaimedTokens`。
 - Codex native compaction 必须直接消费当前 projection 生成的 ResponseItems，不得从 raw branch 重建输入；native checkpoint 同时保存有界 portable summary 供模型或 Provider 切换。
-- 原生远程压缩限定为 `openai-codex` + `openai-codex-responses` 与 `sub-api` + `openai-responses`，默认在 turn boundary 达到 effective 下界 90% 时触发。`src/native-provider.ts` 统一资格与传输选择：Codex 保留原 endpoint/JWT/account ID；SubAPI 从显式 `baseUrl` 解析公共 `/responses`，使用配置的 API key 和请求头，不解析 ChatGPT account ID、不硬编码网关域名，也不调用独立 `/responses/compact`。两者复用 projection、`compaction_trigger`、SSE 结果、native checkpoint 恢复和 portable 模型切换摘要；SubAPI 部署须支持原生 compaction v2。普通聊天支持不代表压缩支持，失败回退 Pi，不自动改写用户模型配置。
+- 原生远程压缩限定为 `openai-codex` + `openai-codex-responses` 与 `sub-api` + `openai-responses`，默认由 L3 阈值或 Pi 自身阈值触发。`src/native-provider.ts` 统一资格与传输选择：Codex 保留原 endpoint/JWT/account ID；SubAPI 从显式 `baseUrl` 解析公共 `/responses`，使用配置的 API key 和请求头，不解析 ChatGPT account ID、不硬编码网关域名，也不调用独立 `/responses/compact`。两者复用 projection、`compaction_trigger`、SSE 结果、native checkpoint 恢复和 portable 模型切换摘要；SubAPI 部署须支持原生 compaction v2。普通聊天支持不代表压缩支持，失败回退 Pi，不自动改写用户模型配置。
 - SubAPI 原生压缩认证必须保留非空显式 `Authorization`（大小写不敏感）的优先级；没有该头时才使用非空 API key 生成 Bearer，均缺失时可接受非空 `cf-aig-authorization` 代理认证头。不得要求 header-only 配置额外提供假 API key，也不得生成 `Bearer undefined`；空白或缺失凭据在网络请求前拒绝并回退 Pi。Codex 仍必须提供有效 JWT，不能用请求头绕过。
 - 同步 Codex 协议代码必须从官方 `ogulcancelik/pi-extensions` 明确 commit 移植并保留对应 MIT 许可证；不得从 npm tarball、缓存、`node_modules` 或构建产物导入。
 
 ## 验证
 
 - 在本目录内至少运行 `npm test`、`npm run check` 和 `npm pack --dry-run --json`；涉及依赖、加载入口、context writer、footer 或 observer 时还要验证根 `npm install` 幂等、锁文件仅包含预期本地 link 变化、根扩展联合加载、性能回归，以及实际 packed 根 tarball 的隔离安装/加载。
-- 修改 projection、token accounting、threshold、checkpoint、模型切换、后台并发或恢复索引后，至少运行 `npm test`、`npm run check`、`npm pack --dry-run --json`、根扩展联合加载、packed 根 tarball 隔离加载和性能回归。性能门禁为 cache-hit P95 <2ms、delta projection P95 <10ms，普通 hot path 不得扫描完整 branch。
+- 修改 projection、token accounting、遮蔽规划/台账、threshold、checkpoint、模型切换或恢复索引后，至少运行 `npm test`、`npm run check`、`npm pack --dry-run --json`、根扩展联合加载、packed 根 tarball 隔离加载和性能回归。性能门禁为 cache-hit P95 <2ms、delta projection P95 <10ms，普通 hot path 不得扫描完整 branch。

pi-compaction/README.md

@@ -1,19 +1,22 @@
 # pi-compaction
 
-High-performance unified context projection and compaction for Pi.
+Pressure-aware context masking and threshold compaction for Pi.
 
 ## Responsibilities
 
-- incrementally index completed tool calls;
+Pi context management is layered; this extension owns the middle two layers:
-- replace large tool results with deterministic bounded evidence stubs;
-- pass small tool results through unchanged whenever a stub would not be shorter;
-- refine large stubs with non-blocking semantic summaries;
-- expose one effective context projection to normal provider requests and Codex/SubAPI native compaction;
-- cancel Pi's raw threshold compaction until the effective lower-bound reaches the configured threshold;
-- retain original tool evidence through `context_tree_query`;
-- publish read-only effective metrics for the footer and context observer.
 
-Context Mode remains responsible for preventing large external data from entering the session. RTK remains responsible for ordinary tool-output cleanup.
+- **L1 — entry truncation** (future `pi-overflow`): truncate new tool results as they are written. Not implemented here.
+- **L2 — observation masking** (this extension): under real context pressure, replace old tool results outside a protected recent window with deterministic one-line placeholders via Pi `context_edit` entries. Originals stay in the session untouched and are recoverable with `context_tree_query`.
+- **L3 — whole-context summary** (this extension): when the session idles (`agent_settled`) above the summary threshold, trigger `ctx.compact` once per leaf.
+- **Pi native threshold compaction** is the final safety net and is never blocked; this extension no longer cancels `session_before_compact` based on its own thresholds.
+
+Additionally it:
+
+- publishes one effective-context snapshot for normal provider requests and Codex/SubAPI native compaction;
+- maintains a masked-results ledger so `/context`, `/compaction-status`, the footer, and AutoReview see the same placeholders the model sees.
+
+pi-overflow bounds new tool results as they arrive. RTK remains responsible for ordinary tool-output cleanup.
 
 ## Configuration
 
@@ -22,21 +25,21 @@ Configuration lives in the installed bundle's `config/pi-compaction.json` (the b
 ```json
 {
   "enabled": true,
-  "softRatio": 0.72,
+  "masking": {
-  "hardRatio": 0.9,
-  "targetRatio": 0.55,
-  "deterministic": {
     "enabled": true,
-    "maxStubTokens": 1024
+    "triggerTokens": 200000,
+    "triggerRatio": 0.6,
+    "targetTokens": 100000,
+    "targetRatio": 0.35,
+    "minClearTokens": 50000,
+    "keepRecentTurns": 10,
+    "minEntryTokens": 256,
+    "excludeTools": ["ask_user_question"]
   },
-  "semantic": {
+  "summary": {
-    "enabled": true,
+    "triggerTokens": 400000,
-    "model": "default",
+    "triggerRatio": 0.9,
-    "thinking": "low",
+    "thinking": "low"
-    "concurrency": 1,
-    "minInputChars": 4000,
-    "maxInputChars": 24000,
-    "maxRetries": 2
   },
   "native": {
     "enabled": true
@@ -44,7 +47,9 @@ Configuration lives in the installed bundle's `config/pi-compaction.json` (the b
 }
 ```
 
-Invalid or missing fields are normalized in memory.
+All thresholds are absolute token counts constrained by a context-window ratio: `effectiveThreshold = min(tokens, floor(ratio × contextWindow))`. Masking fires when `currentTokens ≥ triggerEff` and clears the oldest eligible results until `currentTokens - cleared ≤ targetEff`; if the clearable volume stays under `min(minClearTokens, triggerEff - targetEff)` nothing is masked, so masking is rare, batched, and hysteretic. `masking.target` is normalized below `masking.trigger` and `summary.trigger` above `masking.trigger` when a config would violate that ordering. Invalid or missing fields fall back to defaults in memory.
+
+Masking never touches the most recent `keepRecentTurns` assistant turns, user/assistant messages, `excludeTools`, `SKILL.md` reads, entries already masked, or results smaller than `minEntryTokens`.
 
 ## Native providers
 
@@ -67,17 +72,17 @@ The gateway must have a compatible deployed version, model and available upstrea
 
 ## Session format
 
-The only custom entry type is `pi-compaction`, version 1. Segment summary entries contain hashes, summaries, and source references; they never duplicate complete raw tool output. Native checkpoints use Pi compaction-entry details with the same kind/version.
+The extension writes no `segment-summary` or other custom entries for masking: L2 uses Pi's native `context_edit` entries (`{ type: "context_edit", targetId, replacement }`), which alter only future model context. The raw session entries — and therefore UI, exports, and the evidence ledger — are unchanged; every masked result remains recoverable through `context_tree_query`. Masking placeholders are deterministic single lines starting with `[pi-compaction] Earlier `.
 
-Projection replaces a tool result only when the resulting summary plus its `Source:` recovery line is strictly shorter than the original text content. Small tool results are therefore passed through byte-for-byte and keep their original content parts, including any non-text parts. Because a stub is only larger than the text it would replace when the original is short, this comparison is the sole gate; no separate size option exists. Every indexed result remains recoverable through `context_tree_query` whether or not the message was replaced.
+Compaction entries still carry `pi-compaction` version 1 details; native checkpoints use the same kind/version. This implementation intentionally does not read legacy `context-prune-*`, `openai-codex-native-compaction`, or older `segment-summary` entries. Start a new session after switching to this extension.
-
-This implementation intentionally does not read legacy `context-prune-*` or `openai-codex-native-compaction` entries. Start a new session after switching to this extension.
 
 ## Interfaces
 
-- `context_tree_query`: recover original bounded tool evidence by source key or tool call ID.
+- `context_tree_query`: recover the original output of a tool result that pi-compaction cleared from context, by tool call ID (1–20 ids per call, bounded output). Use it when a `[pi-compaction] Earlier` placeholder lacks evidence needed for the current task.
-- `/compaction-status`: show raw/effective tokens, reclaimed tokens, confidence, segments, and pending semantic jobs.
+- pi-overflow linkage: when the masked tool result carries `details.overflow.id` (pi-overflow stored the oversized original), the placeholder ends with `Full original stored: overflow_query {"id":"…"}`. pi-compaction only reads that public field and never imports pi-overflow internals.
+- `/compaction-status`: show raw/effective tokens, reclaimed tokens, confidence, segments, masked results, and generation.
+- Service `my-pi.pi-compaction.service.v1`: `getSnapshot`, `getMetrics`, and `query(sessionId, ids)` — `query` returns a placeholder record only for tool call IDs that are currently masked, so consumers such as the permission system's AutoReview see the same evidence the model does.
 
 ## Performance model
 
-The first session hydration scans the branch once. Appended turns use delta projection and immutable cached snapshots. The `context` hook performs no model call and no session write. Semantic summaries run in a bounded background worker and become visible only at later safe event boundaries.
+The first session hydration scans the branch once. Appended turns use delta projection and immutable cached snapshots; the `context` hook performs no model call, no session write, and returns the message list unchanged except on the native-checkpoint path, where post-checkpoint `context_edit` entries are applied during projection. Masking drafts are produced in `turn_end` in one batch and only when both the trigger and the `minClear` hysteresis pass, so the hot path stays allocation-free for unprompted turns.

pi-compaction/UPSTREAM.md

@@ -29,3 +29,8 @@
 - 协议证据：`backend/internal/service/openai_compact_body_signal.go` 指出原生远端压缩 v2 为 Responses 加 `stream: true` 与 `compaction_trigger`，并处理 Codex beta 特性。
 - 本地 `src/native-provider.ts` 适配器是针对该协议独立实现的。没有导入任何 Sub2API 源码；复用现有 Codex 序列化器、流解析器和 checkpoint 格式。
 - 该参考描述的是上游能力，不构成对任何特定部署、模型或账号的保证。本适配器不使用单独的 Responses compact 端点。
+
+## 本地改动记录
+
+- 上下文管理重构为 L2/L3：L2 在上下文压力下经 `turn_end` 提交 Pi 原生 `context_edit`，把保护窗口外的旧 toolResult 批量遮蔽为指向 `context_tree_query` 的确定性占位符（原始条目保留、不写 custom entry）；L3 只在空闲 `agent_settled` 且达到绝对 token 阈值（受窗口比例约束）时调用一次 `ctx.compact`。删除逐条确定性存根、后台语义 worker、`softRatio`/`hardRatio`/`targetRatio` 配置与 `session_before_compact` 的自行阈值取消；Pi 原生阈值压缩保留为最终安全网。
+- L2 遮蔽占位符联动 pi-overflow：被遮蔽的 toolResult 若携带 `details.overflow.id`，占位符末尾追加 `overflow_query {"id":"…"}` 恢复提示。

pi-compaction/index.ts

@@ -14,6 +14,12 @@ import {
   type SessionEntry,
 } from "@earendil-works/pi-coding-agent";
 import { DEFAULT_CONFIG, loadConfig } from "./src/config.js";
+import {
+  applyContextEdits,
+  planMasking,
+  rebuildMaskedLedger,
+  type MaskedLedgerEntry,
+} from "./src/masking.js";
 import { buildNativeCompactionTransport, isNativeCompactionModel } from "./src/native-provider.js";
 import {
   buildCompactionRequestBody,
@@ -32,27 +38,28 @@ import {
   type NativeCompactionDetails,
   type ResponseItem,
 } from "./src/native-compaction.js";
-import { SegmentIndex } from "./src/segment-index.js";
+import { contentText, SegmentIndex } from "./src/segment-index.js";
-import { SemanticWorker } from "./src/semantic-worker.js";
-import { shouldCompact } from "./src/token-accounting.js";
 import {
   CompactionServiceRegistry,
   publishCompactionService,
   unpublishCompactionService,
 } from "./src/service.js";
+import { effectiveThreshold } from "./src/token-accounting.js";
 import {
   COMPACTION_CUSTOM_TYPE,
   COMPACTION_VERSION,
   QUERY_TOOL_NAME,
+  type MaskedResultRecord,
   type PiCompactionConfig,
   type ProjectionSnapshot,
-  type ToolHistoryRecord,
 } from "./src/types.js";
 
 interface SessionRuntime {
   index: SegmentIndex;
   config: PiCompactionConfig;
   codexCache?: { fingerprint: string; modelKey: string; toolsHash: string; items: ResponseItem[] };
+  masked: Map<string, MaskedLedgerEntry>;
+  lastAutoCompactLeafId: string | null;
 }
 
 interface CachedPayload {
@@ -64,10 +71,6 @@ function errorMessage(error: unknown): string {
   return error instanceof Error ? error.message : String(error);
 }
 
-function hash(value: string): string {
-  return createHash("sha256").update(value).digest("hex").slice(0, 20);
-}
-
 interface EnvironmentValue {
   hash: string;
   tokens: number;
@@ -81,6 +84,10 @@ interface EnvironmentCacheEntry extends EnvironmentValue {
   tools: unknown[];
 }
 
+function hash(value: string): string {
+  return createHash("sha256").update(value).digest("hex").slice(0, 20);
+}
+
 function environment(
   pi: ExtensionAPI,
   ctx: ExtensionContext,
@@ -182,34 +189,46 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
   const runtimes = new Map<string, SessionRuntime>();
   const payloads = new Map<string, CachedPayload>();
   const environments = new Map<string, EnvironmentCacheEntry>();
-  const worker = new SemanticWorker();
   const service = new CompactionServiceRegistry();
   publishCompactionService(service);
 
+  const maskedRecords = (sessionId: string, ids: string[]): MaskedResultRecord[] => {
+    const masked = runtimes.get(sessionId)?.masked;
+    if (!masked) return [];
+    return ids.flatMap((id) => {
+      const entry = masked.get(id);
+      return entry
+        ? [{
+          toolCallId: id,
+          resultTimestamp: entry.resultTimestamp,
+          deterministicSummary: entry.placeholder,
+        }]
+        : [];
+    });
+  };
+
   const runtimeFor = (ctx: ExtensionContext): SessionRuntime => {
     const sessionId = ctx.sessionManager.getSessionId();
     let runtime = runtimes.get(sessionId);
     if (!runtime) {
-      const index = new SegmentIndex(sessionId);
+      runtime = {
-      index.hydrate(ctx.sessionManager.getBranch() as SessionEntry[], DEFAULT_CONFIG);
+        index: new SegmentIndex(sessionId),
-      runtime = { index, config: structuredClone(DEFAULT_CONFIG) };
+        config: structuredClone(DEFAULT_CONFIG),
+        masked: rebuildLedger(ctx),
+        lastAutoCompactLeafId: null,
+      };
       runtimes.set(sessionId, runtime);
-      service.setSession(sessionId, (ids) => index.query(ids));
+      service.setSession(sessionId, (ids) => maskedRecords(sessionId, ids));
     }
     return runtime;
   };
 
-  const persist = (record: unknown): void => {
+  const ledgerStats = (runtime: SessionRuntime): { maskedResults: number; reclaimedTokens: number } => {
-    pi.appendEntry(COMPACTION_CUSTOM_TYPE, record);
+    let reclaimedTokens = 0;
-  };
+    for (const entry of runtime.masked.values()) {
-
+      reclaimedTokens += Math.max(0, entry.originalTokens - entry.placeholderTokens);
-  const drainSemantic = (ctx: ExtensionContext, runtime: SessionRuntime): void => {
-    const sessionId = ctx.sessionManager.getSessionId();
-    for (const result of worker.drain(sessionId)) {
-      const record = runtime.index.applySemantic(result.sourceKey, result.sourceHash, result.summary);
-      if (record) persist(record);
     }
-    runtime.index.setSemanticJobs(worker.count(sessionId));
+    return { maskedResults: runtime.masked.size, reclaimedTokens };
   };
 
   const projectMessages = (
@@ -218,16 +237,32 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
     messages: readonly any[],
   ): ProjectionSnapshot => {
     const env = environment(pi, ctx, environments);
-    runtime.index.setSemanticJobs(worker.count(ctx.sessionManager.getSessionId()));
+    const stats = ledgerStats(runtime);
     const snapshot = runtime.index.project(messages, {
       environmentHash: env.hash,
       environmentTokens: env.tokens,
       contextWindow: env.contextWindow,
+      maskedResults: stats.maskedResults,
+      reclaimedTokens: stats.reclaimedTokens,
     });
     service.setSnapshot(snapshot);
     return snapshot;
   };
 
+  const rebuildLedger = (ctx: ExtensionContext): Map<string, MaskedLedgerEntry> =>
+    rebuildMaskedLedger(
+      ctx.sessionManager.getBranch(),
+      ctx.sessionManager.getLeafId(),
+      (message) => estimateTokens(message as any),
+    );
+
+  const hydrateRuntime = (ctx: ExtensionContext, runtime: SessionRuntime): void => {
+    runtime.index.hydrate();
+    runtime.masked = rebuildLedger(ctx);
+    runtime.lastAutoCompactLeafId = null;
+    runtime.codexCache = undefined;
+  };
+
   const branchMessages = (
     ctx: ExtensionContext,
     runtime: SessionRuntime,
@@ -237,9 +272,10 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
     const checkpoint = findNativeCheckpoint(branch);
     let messages: any[];
     if (checkpoint.status === "valid") {
-      messages = branch
+      // sessionEntryToContextMessages sees only raw entries; applyContextEdits
-        .slice(checkpoint.checkpoint.entryIndex + 1)
+      // additionally replays the latest context_edit per target (omissions and
-        .flatMap((entry) => sessionEntryToContextMessages(entry));
+      // placeholder replacements included).
+      messages = applyContextEdits(branch.slice(checkpoint.checkpoint.entryIndex + 1));
       if (checkpoint.checkpoint.details.modelKey !== (ctx.model ? modelKey(ctx.model) : "")) {
         messages = [asPortableMessage(checkpoint.checkpoint.details.portableSummary), ...messages];
       }
@@ -278,15 +314,22 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
     publishCompactionService(service);
     const sessionId = ctx.sessionManager.getSessionId();
     const config = await loadConfig();
-    const index = new SegmentIndex(sessionId);
+    const runtime = runtimes.get(sessionId) ?? {
-    index.hydrate(ctx.sessionManager.getBranch() as SessionEntry[], config);
+      index: new SegmentIndex(sessionId),
-    runtimes.set(sessionId, { index, config });
+      config,
-    service.setSession(sessionId, (ids) => index.query(ids));
+      masked: new Map<string, MaskedLedgerEntry>(),
+      lastAutoCompactLeafId: null,
+    };
+    runtime.config = config;
+    runtime.index.hydrate();
+    runtime.masked = rebuildLedger(ctx);
+    runtime.lastAutoCompactLeafId = null;
+    runtimes.set(sessionId, runtime);
+    service.setSession(sessionId, (ids) => maskedRecords(sessionId, ids));
   });
 
   pi.on("session_shutdown", (_event, ctx) => {
     const sessionId = ctx.sessionManager.getSessionId();
-    worker.cancelSession(sessionId);
     runtimes.delete(sessionId);
     payloads.delete(sessionId);
     environments.delete(sessionId);
@@ -296,15 +339,13 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
 
   pi.on("session_tree", (_event, ctx) => {
     const runtime = runtimeFor(ctx);
-    runtime.index.hydrate(ctx.sessionManager.getBranch() as SessionEntry[], runtime.config);
+    hydrateRuntime(ctx, runtime);
-    runtime.codexCache = undefined;
     service.clearSnapshot(ctx.sessionManager.getSessionId());
   });
 
   pi.on("session_compact", (_event, ctx) => {
     const runtime = runtimeFor(ctx);
-    runtime.index.hydrate(ctx.sessionManager.getBranch() as SessionEntry[], runtime.config);
+    hydrateRuntime(ctx, runtime);
-    runtime.codexCache = undefined;
     service.clearSnapshot(ctx.sessionManager.getSessionId());
     payloads.delete(ctx.sessionManager.getSessionId());
   });
@@ -319,21 +360,9 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
 
   pi.on("turn_end", (event, ctx) => {
     const runtime = runtimeFor(ctx);
-    drainSemantic(ctx, runtime);
+    if (!runtime.config.enabled) return undefined;
-    const previous = runtime.index.getSnapshot();
-    const captured = runtime.index.captureTurn(event.message, event.toolResults, runtime.config);
-    for (const record of captured) persist(runtime.index.toPersisted(record));
-    const pressure = previous && previous.contextWindow > 0
-      ? previous.upperBound / previous.contextWindow >= runtime.config.softRatio
-      : false;
-    const semanticCandidates = pressure
-      ? runtime.index.allRecords()
-      : captured.filter((record) => record.rawText.length >= runtime.config.semantic.maxInputChars);
-    for (const record of semanticCandidates) {
-      worker.enqueue(ctx.sessionManager.getSessionId(), record, ctx, runtime.config);
-    }
-    runtime.index.setSemanticJobs(worker.count(ctx.sessionManager.getSessionId()));
 
+    const previous = runtime.index.getSnapshot();
     const usage = ctx.getContextUsage();
     if (usage?.tokens && previous) {
       const env = environment(pi, ctx, environments);
@@ -343,6 +372,50 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
         env.hash,
       );
     }
+
+    const currentTokens = usage?.tokens
+      ?? (event.context?.contextMessages ?? []).reduce((sum, message) => sum + estimateTokens(message as any), 0)
+        + environment(pi, ctx, environments).tokens;
+    const plan = planMasking({
+      contextEntries: event.context?.contextEntries ?? [],
+      currentTokens,
+      contextWindow: ctx.model?.contextWindow ?? 0,
+      masking: runtime.config.masking,
+      estimate: (message) => estimateTokens(message as any),
+    });
+    if (!plan.drafts.length) return undefined;
+    for (const result of plan.masked) {
+      runtime.masked.set(result.toolCallId, {
+        placeholder: result.placeholder,
+        originalTokens: result.originalTokens,
+        placeholderTokens: result.placeholderTokens,
+        resultTimestamp: result.resultTimestamp,
+      });
+    }
+    return { entries: [...(event.entries ?? []), ...plan.drafts] };
+  });
+
+  pi.on("agent_settled", (_event, ctx) => {
+    const runtime = runtimeFor(ctx);
+    if (!runtime.config.enabled || ctx.isIdle?.() === false) return;
+    const snapshot = runtime.index.getSnapshot();
+    const tokens = ctx.getContextUsage()?.tokens ?? snapshot?.effectiveTokens;
+    if (tokens == null) return;
+    const contextWindow = ctx.model?.contextWindow ?? snapshot?.contextWindow ?? 0;
+    const threshold = effectiveThreshold(
+      runtime.config.summary.triggerTokens,
+      runtime.config.summary.triggerRatio,
+      contextWindow,
+    );
+    if (tokens < threshold) return;
+    const leafId = ctx.sessionManager.getLeafId();
+    if (leafId !== null && runtime.lastAutoCompactLeafId === leafId) return;
+    runtime.lastAutoCompactLeafId = leafId;
+    ctx.compact({
+      onError: (error) => {
+        if (ctx.hasUI) ctx.ui.notify(`pi-compaction auto summary failed: ${errorMessage(error)}`, "warning");
+      },
+    });
   });
 
   pi.on("context", (event, ctx) => {
@@ -358,7 +431,10 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
       const projected = branchMessages(ctx, runtime, branch);
       return { messages: projected.snapshot.messages as any };
     }
-    return { messages: projectMessages(ctx, runtime, event.messages).messages as any };
+    // Non-checkpoint path only refreshes the shared snapshot; the message list
+    // is returned to Pi unchanged — masking is append-only via context_edit.
+    projectMessages(ctx, runtime, event.messages);
+    return undefined;
   });
 
   pi.on("before_provider_headers", (event, ctx) => {
@@ -387,7 +463,6 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
   pi.on("session_before_compact", async (event, ctx) => {
     const runtime = runtimeFor(ctx);
     if (!runtime.config.enabled) return undefined;
-    drainSemantic(ctx, runtime);
     const { snapshot, checkpoint } = branchMessages(
       ctx,
       runtime,
@@ -395,8 +470,6 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
       event.reason === "overflow" && event.willRetry,
     );
 
-    if (!shouldCompact(event.reason, snapshot, runtime.config)) return { cancel: true };
-
     const model = ctx.model;
     if (!model) return undefined;
     if (!runtime.config.native.enabled || !isNativeCompactionModel(model)) {
@@ -412,7 +485,7 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
           event.signal,
           event.customInstructions,
           undefined,
-          runtime.config.semantic.thinking,
+          runtime.config.summary.thinking,
         );
         return {
           compaction: {
@@ -463,7 +536,7 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
         replacementHistory: buildReplacementHistory(
           input,
           remote.compactionItem,
-          Math.min(64_000, Math.max(4_096, Math.floor(snapshot.contextWindow * Math.max(0.05, runtime.config.targetRatio - 0.3)))),
+          Math.min(64_000, Math.max(4_096, Math.floor(snapshot.contextWindow * 0.25))),
         ),
         portableSummary: portableSummary(snapshot.messages),
         sourceFingerprint: snapshot.fingerprint,
@@ -504,7 +577,7 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
         event.signal,
         event.preparation.customInstructions,
         undefined,
-        runtime.config.semantic.thinking,
+        runtime.config.summary.thinking,
         undefined,
         undefined,
         undefined,
@@ -531,38 +604,64 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
   pi.registerTool({
     name: QUERY_TOOL_NAME,
     label: "Query Compacted Tool History",
-    description: "Recover original tool outputs by source key or tool call id after pi-compaction replaced them with bounded summaries.",
+    description: "Recover the original output of a tool result that pi-compaction cleared from context, by tool call id.",
-    promptSnippet: "Recover original compacted tool evidence",
+    promptSnippet: "Recover original tool output cleared from context",
     promptGuidelines: [
-      "Use context_tree_query only when a pi-compaction summary lacks evidence needed for the current task.",
+      "Use context_tree_query when you see a \"[pi-compaction] Earlier\" placeholder and still need the original tool output.",
     ],
     parameters: Type.Object({
       toolCallIds: Type.Array(Type.String(), { minItems: 1, maxItems: 20 }),
     }),
     async execute(_id, params, _signal, _update, ctx) {
-      const records = service.query(ctx.sessionManager.getSessionId(), params.toolCallIds);
+      const branch = ctx.sessionManager.getBranch() as SessionEntry[];
-      const blocks = records.map((record: ToolHistoryRecord) => {
+      const calls = new Map<string, { toolName: string; args: unknown }>();
-        const perResultBytes = Math.max(1_024, Math.floor(DEFAULT_MAX_BYTES / Math.max(1, records.length)));
+      const results = new Map<string, any>();
-        const args = truncateHead(JSON.stringify(record.args), { maxLines: 20, maxBytes: 2_048 }).content;
+      for (const entry of branch) {
-        const truncated = truncateHead(record.rawText, {
+        const message = (entry as any)?.message;
-          maxLines: Math.max(20, Math.floor(DEFAULT_MAX_LINES / Math.max(1, records.length))),
+        if ((entry as any)?.type !== "message" || !message) continue;
+        if (message.role === "assistant" && Array.isArray(message.content)) {
+          for (const block of message.content) {
+            if (block?.type === "toolCall" && typeof block.id === "string") {
+              calls.set(block.id, {
+                toolName: String(block.name ?? "unknown"),
+                args: block.arguments ?? block.input ?? block.args ?? {},
+              });
+            }
+          }
+        } else if (message.role === "toolResult" && typeof message.toolCallId === "string") {
+          results.set(message.toolCallId, message);
+        }
+      }
+
+      const found: string[] = [];
+      const blocks: string[] = [];
+      for (const id of params.toolCallIds) {
+        const result = results.get(id);
+        if (!result) continue;
+        found.push(id);
+        const call = calls.get(id) ?? { toolName: String(result.toolName ?? "unknown"), args: {} };
+        const perResultBytes = Math.max(1_024, Math.floor(DEFAULT_MAX_BYTES / Math.max(1, params.toolCallIds.length)));
+        const args = truncateHead(JSON.stringify(call.args ?? {}), { maxLines: 20, maxBytes: 2_048 }).content;
+        const rawText = contentText(result.content);
+        const truncated = truncateHead(rawText, {
+          maxLines: Math.max(20, Math.floor(DEFAULT_MAX_LINES / Math.max(1, params.toolCallIds.length))),
           maxBytes: perResultBytes,
         });
-        return [
+        blocks.push([
-          `## ${record.sourceKey}`,
+          `## ${id}`,
-          `Tool: ${record.toolName}`,
+          `Tool: ${call.toolName}`,
           `Args: ${args}`,
-          `Status: ${record.isError ? "ERROR" : "OK"}`,
+          `Status: ${result.isError ? "ERROR" : "OK"}`,
           "",
           truncated.content,
           ...(truncated.truncated ? [`[truncated: ${truncated.outputLines}/${truncated.totalLines} lines]`] : []),
-        ].join("\n");
+        ].join("\n"));
-      });
+      }
-      const missing = params.toolCallIds.filter((id: string) => !records.some((record) => record.sourceKey === id || record.toolCallId === id));
+      const missing = params.toolCallIds.filter((id: string) => !results.has(id));
       for (const id of missing) blocks.push(`## ${id}\n(not found)`);
       return {
         content: [{ type: "text", text: blocks.join("\n\n---\n\n") }],
-        details: { sourceKeys: records.map((record) => record.sourceKey), missing },
+        details: { toolCallIds: found, missing },
       };
     },
   });
@@ -576,12 +675,11 @@ export default function piCompactionExtension(pi: ExtensionAPI): void {
             `Effective: ${metrics.effectiveTokens}/${metrics.contextWindow} (${metrics.percent.toFixed(1)}%, ${metrics.confidence})`,
             `Raw: ${metrics.rawTokens}/${metrics.contextWindow}`,
             `Reclaimed: ${metrics.reclaimedTokens}`,
-            `Segments: ${metrics.segmentCount}; semantic jobs: ${metrics.semanticJobs}; generation: ${metrics.generation}`,
+            `Segments: ${metrics.segmentCount}; masked tool results: ${metrics.maskedResults}; generation: ${metrics.generation}`,
           ].join("\n")
         : "pi-compaction has no projection snapshot yet.";
       if (ctx.hasUI) ctx.ui.notify(text, "info");
       else process.stderr.write(`${text}\n`);
     },
   });
-
 }

pi-compaction/src/config.ts

@@ -12,27 +12,29 @@ export const CONFIG_PATH = fileURLToPath(
 
 export const DEFAULT_CONFIG: PiCompactionConfig = {
   enabled: true,
-  softRatio: 0.72,
+  masking: {
-  hardRatio: 0.9,
-  targetRatio: 0.55,
-  deterministic: {
     enabled: true,
-    maxStubTokens: 1_024,
+    triggerTokens: 200_000,
+    triggerRatio: 0.6,
+    targetTokens: 100_000,
+    targetRatio: 0.35,
+    minClearTokens: 50_000,
+    keepRecentTurns: 10,
+    minEntryTokens: 256,
+    excludeTools: ["ask_user_question"],
   },
-  semantic: {
+  summary: {
-    enabled: true,
+    triggerTokens: 400_000,
-    model: "default",
+    triggerRatio: 0.9,
     thinking: "low",
-    concurrency: 1,
-    minInputChars: 4_000,
-    maxInputChars: 24_000,
-    maxRetries: 2,
   },
   native: {
     enabled: true,
   },
 };
 
+const THINKING_LEVELS = ["off", "minimal", "low", "medium", "high", "xhigh"];
+
 function ratio(value: unknown, fallback: number): number {
   return typeof value === "number" && Number.isFinite(value) && value > 0 && value < 1 ? value : fallback;
 }
@@ -41,46 +43,67 @@ function integer(value: unknown, fallback: number, min = 0, max = Number.MAX_SAF
   return typeof value === "number" && Number.isInteger(value) && value >= min && value <= max ? value : fallback;
 }
 
+function positiveInteger(value: unknown, fallback: number): number {
+  return integer(value, fallback, 1);
+}
+
 function object(value: unknown): Record<string, unknown> {
   return value !== null && typeof value === "object" && !Array.isArray(value)
     ? value as Record<string, unknown>
     : {};
 }
 
+function stringArray(value: unknown, fallback: string[]): string[] {
+  if (!Array.isArray(value) || !value.every((entry) => typeof entry === "string")) return fallback;
+  return value.map((entry) => entry.trim()).filter((entry) => entry.length > 0);
+}
+
 export function normalizeConfig(value: unknown): PiCompactionConfig {
   const input = object(value);
-  const deterministic = object(input.deterministic);
+  const masking = object(input.masking);
-  const semantic = object(input.semantic);
+  const summary = object(input.summary);
   const native = object(input.native);
-  const softRatio = ratio(input.softRatio, DEFAULT_CONFIG.softRatio);
-  const hardRatio = Math.max(softRatio + 0.01, ratio(input.hardRatio, DEFAULT_CONFIG.hardRatio));
-  const targetRatio = Math.min(softRatio - 0.01, ratio(input.targetRatio, DEFAULT_CONFIG.targetRatio));
-  const thinking = ["off", "minimal", "low", "medium", "high", "xhigh"].includes(String(semantic.thinking))
-    ? semantic.thinking as PiCompactionConfig["semantic"]["thinking"]
-    : DEFAULT_CONFIG.semantic.thinking;
 
-  return {
+  const normalized: PiCompactionConfig = {
     enabled: typeof input.enabled === "boolean" ? input.enabled : DEFAULT_CONFIG.enabled,
-    softRatio,
+    masking: {
-    hardRatio: Math.min(0.99, hardRatio),
+      enabled: typeof masking.enabled === "boolean" ? masking.enabled : DEFAULT_CONFIG.masking.enabled,
-    targetRatio: Math.max(0.1, targetRatio),
+      triggerTokens: positiveInteger(masking.triggerTokens, DEFAULT_CONFIG.masking.triggerTokens),
-    deterministic: {
+      triggerRatio: ratio(masking.triggerRatio, DEFAULT_CONFIG.masking.triggerRatio),
-      enabled: typeof deterministic.enabled === "boolean" ? deterministic.enabled : DEFAULT_CONFIG.deterministic.enabled,
+      targetTokens: positiveInteger(masking.targetTokens, DEFAULT_CONFIG.masking.targetTokens),
-      maxStubTokens: integer(deterministic.maxStubTokens, DEFAULT_CONFIG.deterministic.maxStubTokens, 32, 4_096),
+      targetRatio: ratio(masking.targetRatio, DEFAULT_CONFIG.masking.targetRatio),
+      minClearTokens: positiveInteger(masking.minClearTokens, DEFAULT_CONFIG.masking.minClearTokens),
+      keepRecentTurns: integer(masking.keepRecentTurns, DEFAULT_CONFIG.masking.keepRecentTurns, 1, 100),
+      minEntryTokens: integer(masking.minEntryTokens, DEFAULT_CONFIG.masking.minEntryTokens, 0, 100_000),
+      excludeTools: stringArray(masking.excludeTools, DEFAULT_CONFIG.masking.excludeTools),
     },
-    semantic: {
+    summary: {
-      enabled: typeof semantic.enabled === "boolean" ? semantic.enabled : DEFAULT_CONFIG.semantic.enabled,
+      triggerTokens: positiveInteger(summary.triggerTokens, DEFAULT_CONFIG.summary.triggerTokens),
-      model: typeof semantic.model === "string" && semantic.model.trim() ? semantic.model.trim() : DEFAULT_CONFIG.semantic.model,
+      triggerRatio: ratio(summary.triggerRatio, DEFAULT_CONFIG.summary.triggerRatio),
-      thinking,
+      thinking: THINKING_LEVELS.includes(String(summary.thinking))
-      concurrency: integer(semantic.concurrency, DEFAULT_CONFIG.semantic.concurrency, 1, 4),
+        ? summary.thinking as PiCompactionConfig["summary"]["thinking"]
-      minInputChars: integer(semantic.minInputChars, DEFAULT_CONFIG.semantic.minInputChars, 0, 1_000_000),
+        : DEFAULT_CONFIG.summary.thinking,
-      maxInputChars: integer(semantic.maxInputChars, DEFAULT_CONFIG.semantic.maxInputChars, 1_000, 1_000_000),
-      maxRetries: integer(semantic.maxRetries, DEFAULT_CONFIG.semantic.maxRetries, 0, 5),
     },
     native: {
       enabled: typeof native.enabled === "boolean" ? native.enabled : DEFAULT_CONFIG.native.enabled,
     },
   };
+
+  // Trigger/target must keep their hysteresis ordering; fix up rather than
+  // letting a degenerate pair mask-or-compact every turn.
+  if (normalized.masking.targetTokens >= normalized.masking.triggerTokens) {
+    normalized.masking.targetTokens = Math.floor(normalized.masking.triggerTokens / 2);
+  }
+  if (normalized.masking.targetRatio >= normalized.masking.triggerRatio) {
+    normalized.masking.targetRatio = normalized.masking.triggerRatio / 2;
+  }
+  if (normalized.summary.triggerTokens <= normalized.masking.triggerTokens) {
+    normalized.summary.triggerTokens = normalized.masking.triggerTokens * 2;
+  }
+  if (normalized.summary.triggerRatio <= normalized.masking.triggerRatio) {
+    normalized.summary.triggerRatio = Math.min(0.99, normalized.masking.triggerRatio + 0.01);
+  }
+  return normalized;
 }
 
 export async function loadConfig(

pi-compaction/src/masking.ts

@@ -0,0 +1,257 @@
+import {
+  buildContextEntries,
+  sessionEntryToContextMessages,
+  type ContextEditEntryDraft,
+} from "@earendil-works/pi-coding-agent";
+import { contentText, timestampOf } from "./segment-index.js";
+import { effectiveThreshold } from "./token-accounting.js";
+import type { MaskingConfig } from "./types.js";
+
+// Every placeholder written through a context_edit starts with this prefix so
+// re-masking, ledger rebuilds, and the model-facing recovery hint all key off
+// one marker.
+export const MASK_PREFIX = "[pi-compaction] Earlier ";
+
+const MAX_PLACEHOLDER_ARGS_CHARS = 300;
+const EDITABLE_ROLES = new Set(["user", "assistant", "toolResult", "custom"]);
+
+export interface MaskingProjectedEntry {
+  sourceEntry: { id?: string; type?: string; message?: any };
+  messages: any[];
+}
+
+export interface MaskedResult {
+  toolCallId: string;
+  entryId: string;
+  placeholder: string;
+  originalTokens: number;
+  placeholderTokens: number;
+  resultTimestamp: number;
+}
+
+export interface MaskingPlan {
+  drafts: ContextEditEntryDraft[];
+  clearedTokens: number;
+  masked: MaskedResult[];
+}
+
+export interface MaskedLedgerEntry {
+  placeholder: string;
+  originalTokens: number;
+  placeholderTokens: number;
+  resultTimestamp: number;
+}
+
+const EMPTY_PLAN: MaskingPlan = { drafts: [], clearedTokens: 0, masked: [] };
+
+function json(value: unknown): string {
+  try {
+    return JSON.stringify(value) ?? "undefined";
+  } catch {
+    return "[unserializable]";
+  }
+}
+
+function maskPlaceholder(
+  toolName: string,
+  args: unknown,
+  toolCallId: string,
+  originalTokens: number,
+  overflowId?: string,
+): string {
+  let argsText = json(args);
+  if (argsText.length > MAX_PLACEHOLDER_ARGS_CHARS) {
+    argsText = `${argsText.slice(0, MAX_PLACEHOLDER_ARGS_CHARS)}…`;
+  }
+  // `details.overflow.id` is pi-overflow's public contract — read the field
+  // only, never import its implementation.
+  const overflowHint = typeof overflowId === "string" && overflowId.length > 0
+    ? ` Full original stored: overflow_query {"id":"${overflowId}"}.`
+    : "";
+  return `${MASK_PREFIX}${toolName} result cleared to save context (~${originalTokens} tokens). ` +
+    `Call args: ${argsText}. Re-run the call if still needed, or recover the original with ` +
+    `context_tree_query {"toolCallIds":["${toolCallId}"]}.${overflowHint}`;
+}
+
+function isSkillRead(toolName: string, args: unknown): boolean {
+  if (toolName !== "read" || args === null || typeof args !== "object") return false;
+  const record = args as Record<string, unknown>;
+  const raw = record.path ?? record.filePath ?? record.file;
+  if (typeof raw !== "string") return false;
+  return (raw.split(/[\\/]/).pop() ?? raw) === "SKILL.md";
+}
+
+function toolResultCalls(entries: readonly MaskingProjectedEntry[]): Map<string, { toolName: string; args: unknown }> {
+  const calls = new Map<string, { toolName: string; args: unknown }>();
+  for (const entry of entries) {
+    for (const message of entry.messages ?? []) {
+      if (message?.role !== "assistant" || !Array.isArray(message.content)) continue;
+      for (const block of message.content) {
+        if (block?.type === "toolCall" && typeof block.id === "string") {
+          calls.set(block.id, {
+            toolName: String(block.name ?? "unknown"),
+            args: block.arguments ?? block.input ?? block.args ?? {},
+          });
+        }
+      }
+    }
+  }
+  return calls;
+}
+
+export function planMasking(params: {
+  contextEntries: readonly MaskingProjectedEntry[];
+  currentTokens: number;
+  contextWindow: number;
+  masking: MaskingConfig;
+  estimate: (message: any) => number;
+}): MaskingPlan {
+  const { masking } = params;
+  const triggerEff = effectiveThreshold(masking.triggerTokens, masking.triggerRatio, params.contextWindow);
+  if (!masking.enabled || params.currentTokens < triggerEff) return EMPTY_PLAN;
+  const targetEff = effectiveThreshold(masking.targetTokens, masking.targetRatio, params.contextWindow);
+  const minClear = Math.min(masking.minClearTokens, Math.max(0, triggerEff - targetEff));
+
+  const entries = params.contextEntries;
+  // The last keepRecentTurns assistant messages and everything after them stay
+  // untouched; when fewer turns exist, the whole context is protected.
+  let protectedIndex = 0;
+  let turns = 0;
+  for (let index = entries.length - 1; index >= 0; index--) {
+    const entry = entries[index];
+    if (entry?.sourceEntry?.type === "message" && (entry.messages ?? []).some((m) => m?.role === "assistant")) {
+      turns++;
+      if (turns >= masking.keepRecentTurns) {
+        protectedIndex = index;
+        break;
+      }
+    }
+  }
+
+  const calls = toolResultCalls(entries);
+  const needed = params.currentTokens - targetEff;
+  const masked: MaskedResult[] = [];
+  let clearedTokens = 0;
+  for (let index = 0; index < protectedIndex && clearedTokens < needed; index++) {
+    const entry = entries[index];
+    const entryId = entry?.sourceEntry?.id;
+    if (entry?.sourceEntry?.type !== "message" || typeof entryId !== "string") continue;
+    for (const message of entry.messages ?? []) {
+      if (clearedTokens >= needed) break;
+      if (message?.role !== "toolResult" || typeof message.toolCallId !== "string") continue;
+      const call = calls.get(message.toolCallId);
+      const toolName = call?.toolName ?? String(message.toolName ?? "unknown");
+      if (masking.excludeTools.includes(toolName)) continue;
+      const args = call?.args ?? {};
+      if (isSkillRead(toolName, args)) continue;
+      if (contentText(message.content).startsWith(MASK_PREFIX)) continue;
+      const originalTokens = params.estimate(message);
+      if (originalTokens < masking.minEntryTokens) continue;
+      const overflowId = (message.details as { overflow?: { id?: unknown } } | undefined)?.overflow?.id;
+      const placeholder = maskPlaceholder(
+        toolName,
+        args,
+        message.toolCallId,
+        originalTokens,
+        typeof overflowId === "string" ? overflowId : undefined,
+      );
+      const placeholderTokens = params.estimate({ ...message, content: [{ type: "text", text: placeholder }] });
+      clearedTokens += originalTokens - placeholderTokens;
+      masked.push({
+        toolCallId: message.toolCallId,
+        entryId,
+        placeholder,
+        originalTokens,
+        placeholderTokens,
+        resultTimestamp: timestampOf(message.timestamp),
+      });
+    }
+  }
+  if (clearedTokens < minClear) return EMPTY_PLAN;
+  return {
+    drafts: masked.map((result) => ({
+      type: "context_edit",
+      targetId: result.entryId,
+      replacement: { content: result.placeholder },
+    })),
+    clearedTokens,
+    masked,
+  };
+}
+
+/**
+ * Rebuild the masked-results ledger from a session branch after hydration:
+ * the latest context_edit per target wins, only edits carrying this
+ * extension's placeholder prefix count as masked, and only targets that are
+ * still part of the active context (post-compaction kept set) are counted.
+ */
+export function rebuildMaskedLedger(
+  branch: readonly any[],
+  leafId: string | null | undefined,
+  estimate: (message: any) => number,
+): Map<string, MaskedLedgerEntry> {
+  const latestEdits = new Map<string, any>();
+  const targets = new Map<string, any>();
+  for (const entry of branch) {
+    if (entry?.type === "context_edit" && typeof entry.targetId === "string") {
+      latestEdits.set(entry.targetId, entry);
+    } else if (entry?.type === "message" && typeof entry.id === "string") {
+      targets.set(entry.id, entry);
+    }
+  }
+  const contextIds = new Set(
+    buildContextEntries(branch as any, leafId).map((entry) => entry.id),
+  );
+  const masked = new Map<string, MaskedLedgerEntry>();
+  for (const [targetId, edit] of latestEdits) {
+    if (!contextIds.has(targetId)) continue;
+    const text = contentText(edit.replacement?.content);
+    if (!text.startsWith(MASK_PREFIX)) continue;
+    const message = targets.get(targetId)?.message;
+    if (message?.role !== "toolResult" || typeof message.toolCallId !== "string") continue;
+    masked.set(message.toolCallId, {
+      placeholder: text,
+      originalTokens: estimate(message),
+      placeholderTokens: estimate({ ...message, content: [{ type: "text", text }] }),
+      resultTimestamp: timestampOf(message.timestamp),
+    });
+  }
+  return masked;
+}
+
+/**
+ * Apply the latest context_edit per target the way buildSessionProjection does
+ * for message entries `sessionEntryToContextMessages` alone would miss: null
+ * omits the entry, a string replacement becomes one text block for
+ * assistant/toolResult roles.
+ */
+export function applyContextEdits(entries: readonly any[]): any[] {
+  const edits = new Map<string, { content: unknown } | null>();
+  for (const entry of entries) {
+    if (entry?.type === "context_edit" && typeof entry.targetId === "string") {
+      edits.set(entry.targetId, entry.replacement ?? null);
+    }
+  }
+  const projected: any[] = [];
+  for (const entry of entries) {
+    const messages = sessionEntryToContextMessages(entry);
+    const edit = edits.get(entry?.id);
+    if (edit === undefined) {
+      projected.push(...messages);
+      continue;
+    }
+    if (edit === null) continue;
+    for (const message of messages) {
+      if (!EDITABLE_ROLES.has(message?.role)) {
+        projected.push(message);
+        continue;
+      }
+      const content = (message.role === "assistant" || message.role === "toolResult")
+        && typeof edit.content === "string"
+        ? [{ type: "text", text: edit.content }]
+        : edit.content;
+      projected.push({ ...message, content });
+    }
+  }
+  return projected;
+}

pi-compaction/src/segment-index.ts

@@ -1,16 +1,8 @@
 import { createHash } from "node:crypto";
 import { estimateTokens } from "@earendil-works/pi-coding-agent";
-import {
+import type { ProjectionSnapshot, UsageBaseline } from "./types.js";
-  COMPACTION_CUSTOM_TYPE,
-  COMPACTION_VERSION,
-  type PiCompactionConfig,
-  type ProjectionSnapshot,
-  type SegmentSummaryRecord,
-  type ToolHistoryRecord,
-  type UsageBaseline,
-} from "./types.js";
 
-function timestampOf(value: unknown): number {
+export function timestampOf(value: unknown): number {
   if (typeof value === "number" && Number.isFinite(value)) return value;
   if (typeof value === "string") {
     const parsed = Date.parse(value);
@@ -19,10 +11,6 @@ function timestampOf(value: unknown): number {
   return 0;
 }
 
-export function sourceKey(toolCallId: string, timestamp: unknown): string {
-  return `${toolCallId}@${timestampOf(timestamp)}`;
-}
-
 export function contentText(content: unknown): string {
   if (typeof content === "string") return content;
   if (!Array.isArray(content)) return "";
@@ -37,42 +25,6 @@ function hash(value: string): string {
   return createHash("sha256").update(value).digest("hex").slice(0, 20);
 }
 
-function json(value: unknown): string {
-  try {
-    return JSON.stringify(value) ?? "undefined";
-  } catch {
-    return "[unserializable]";
-  }
-}
-
-function boundedEvidence(text: string, maxChars: number): string {
-  const normalized = text.replace(/\u001b\[[0-9;]*m/g, "").trim();
-  if (normalized.length <= maxChars) return normalized;
-  const marker = `\n… ${normalized.length - maxChars} chars omitted …\n`;
-  const available = Math.max(0, maxChars - marker.length);
-  const head = Math.ceil(available * 0.7);
-  const tail = Math.max(0, available - head);
-  return `${normalized.slice(0, head)}${marker}${tail ? normalized.slice(-tail) : ""}`;
-}
-
-export function deterministicSummary(params: {
-  toolName: string;
-  args: unknown;
-  rawText: string;
-  isError: boolean;
-  maxStubTokens: number;
-}): string {
-  const maxChars = params.maxStubTokens * 4;
-  const argsText = boundedEvidence(json(params.args), Math.min(600, Math.floor(maxChars * 0.25)));
-  const evidenceBudget = Math.max(128, maxChars - argsText.length - 120);
-  const evidence = boundedEvidence(params.rawText || "(no output)", evidenceBudget);
-  return [
-    `[pi-compaction tool=${params.toolName} status=${params.isError ? "error" : "ok"}]`,
-    `Args: ${argsText}`,
-    `Evidence:\n${evidence}`,
-  ].join("\n");
-}
-
 function messageKey(message: any): string {
   const role = String(message?.role ?? "unknown");
   const timestamp = timestampOf(message?.timestamp);
@@ -85,24 +37,14 @@ function messageKey(message: any): string {
   return `${role}:${timestamp}:${typeof message?.content === "string" ? message.content.length : (message?.content?.length ?? 0)}`;
 }
 
-function isSummaryRecord(value: unknown): value is SegmentSummaryRecord {
+/**
-  if (!value || typeof value !== "object") return false;
+ * Incremental context projection: tracks the branch's message list, estimates
-  const record = value as Partial<SegmentSummaryRecord>;
+ * tokens, and maintains the usage-backed effective-token baseline. Projection
-  return record.version === COMPACTION_VERSION
+ * is identity — pi-compaction never rewrites messages on this path; context
-    && record.kind === "segment-summary"
+ * pressure reduction happens through append-only context_edit entries.
-    && typeof record.sourceKey === "string"
+ */
-    && typeof record.toolCallId === "string"
-    && typeof record.resultTimestamp === "number"
-    && typeof record.toolName === "string"
-    && typeof record.sourceHash === "string"
-    && typeof record.deterministicSummary === "string";
-}
-
 export class SegmentIndex {
-  private readonly records = new Map<string, ToolHistoryRecord>();
-  private readonly ids = new Map<string, string[]>();
   private generation = 0;
-  private semanticJobs = 0;
   private usageBaseline?: UsageBaseline;
   private sourceMessages: any[] = [];
   private sourceKeys: string[] = [];
@@ -117,67 +59,13 @@ export class SegmentIndex {
     this.sessionId = sessionId;
   }
 
-  hydrate(branch: readonly any[], config: PiCompactionConfig): void {
+  hydrate(): void {
     this.reset();
-    if (!config.enabled) return;
-    const summaries = new Map<string, SegmentSummaryRecord>();
-    for (const entry of branch) {
-      if (entry?.type === "custom" && entry.customType === COMPACTION_CUSTOM_TYPE && isSummaryRecord(entry.data)) {
-        summaries.set(entry.data.sourceKey, entry.data);
-      }
-    }
-
-    const pendingCalls = new Map<string, { toolName: string; args: unknown }>();
-    for (const entry of branch) {
-      if (entry?.type !== "message") continue;
-      const message = entry.message;
-      if (message?.role === "assistant" && Array.isArray(message.content)) {
-        for (const block of message.content) {
-          if (block?.type === "toolCall" && typeof block.id === "string") {
-            pendingCalls.set(block.id, {
-              toolName: String(block.name ?? "unknown"),
-              args: block.input ?? block.arguments ?? block.args ?? {},
-            });
-          }
-        }
-      } else if (message?.role === "toolResult" && typeof message.toolCallId === "string") {
-        const call = pendingCalls.get(message.toolCallId) ?? {
-          toolName: String(message.toolName ?? "unknown"),
-          args: {},
-        };
-        const key = sourceKey(message.toolCallId, message.timestamp);
-        const rawText = contentText(message.content);
-        const persisted = summaries.get(key);
-        const record: ToolHistoryRecord = {
-          sourceKey: key,
-          toolCallId: message.toolCallId,
-          resultTimestamp: timestampOf(message.timestamp),
-          toolName: call.toolName,
-          args: call.args,
-          isError: Boolean(message.isError),
-          rawText,
-          sourceHash: hash(`${call.toolName}\0${json(call.args)}\0${rawText}`),
-          deterministicSummary: persisted?.deterministicSummary ?? deterministicSummary({
-            toolName: call.toolName,
-            args: call.args,
-            rawText,
-            isError: Boolean(message.isError),
-            maxStubTokens: config.deterministic.maxStubTokens,
-          }),
-          semanticSummary: persisted?.semanticSummary,
-        };
-        this.put(record);
-        pendingCalls.delete(message.toolCallId);
-      }
-    }
     this.generation++;
   }
 
   reset(): void {
-    this.records.clear();
-    this.ids.clear();
     this.generation = 0;
-    this.semanticJobs = 0;
     this.usageBaseline = undefined;
     this.sourceMessages = [];
     this.sourceKeys = [];
@@ -187,86 +75,6 @@ export class SegmentIndex {
     this.cached = undefined;
   }
 
-  private put(record: ToolHistoryRecord): void {
-    this.records.set(record.sourceKey, record);
-    const keys = this.ids.get(record.toolCallId) ?? [];
-    if (!keys.includes(record.sourceKey)) keys.push(record.sourceKey);
-    this.ids.set(record.toolCallId, keys);
-  }
-
-  captureTurn(message: any, toolResults: any[], config: PiCompactionConfig): ToolHistoryRecord[] {
-    if (!config.enabled || !config.deterministic.enabled || !Array.isArray(message?.content)) return [];
-    const calls = new Map<string, any>();
-    for (const block of message.content) {
-      if (block?.type === "toolCall" && typeof block.id === "string") calls.set(block.id, block);
-    }
-    const captured: ToolHistoryRecord[] = [];
-    for (const result of toolResults ?? []) {
-      if (!result || typeof result.toolCallId !== "string") continue;
-      const call = calls.get(result.toolCallId);
-      if (!call) continue;
-      const rawText = contentText(result.content);
-      const toolName = String(call.name ?? result.toolName ?? "unknown");
-      const args = call.input ?? call.arguments ?? call.args ?? {};
-      const key = sourceKey(result.toolCallId, result.timestamp);
-      const record: ToolHistoryRecord = {
-        sourceKey: key,
-        toolCallId: result.toolCallId,
-        resultTimestamp: timestampOf(result.timestamp),
-        toolName,
-        args,
-        isError: Boolean(result.isError),
-        rawText,
-        sourceHash: hash(`${toolName}\0${json(args)}\0${rawText}`),
-        deterministicSummary: deterministicSummary({
-          toolName,
-          args,
-          rawText,
-          isError: Boolean(result.isError),
-          maxStubTokens: config.deterministic.maxStubTokens,
-        }),
-      };
-      this.put(record);
-      captured.push(record);
-    }
-    if (captured.length) {
-      this.generation++;
-      this.cached = undefined;
-    }
-    return captured;
-  }
-
-  applySemantic(sourceKeyValue: string, sourceHash: string, summary: string): SegmentSummaryRecord | undefined {
-    const record = this.records.get(sourceKeyValue);
-    if (!record || record.sourceHash !== sourceHash || !summary.trim()) return undefined;
-    record.semanticSummary = summary.trim();
-    this.generation++;
-    this.cached = undefined;
-    return this.toPersisted(record);
-  }
-
-  toPersisted(record: ToolHistoryRecord): SegmentSummaryRecord {
-    const effective = record.semanticSummary ?? record.deterministicSummary;
-    return {
-      version: COMPACTION_VERSION,
-      kind: "segment-summary",
-      sourceKey: record.sourceKey,
-      toolCallId: record.toolCallId,
-      resultTimestamp: record.resultTimestamp,
-      toolName: record.toolName,
-      sourceHash: record.sourceHash,
-      deterministicSummary: record.deterministicSummary,
-      ...(record.semanticSummary ? { semanticSummary: record.semanticSummary } : {}),
-      rawChars: record.rawText.length,
-      effectiveChars: effective.length,
-    };
-  }
-
-  setSemanticJobs(count: number): void {
-    this.semanticJobs = Math.max(0, count);
-    if (this.cached) this.cached = { ...this.cached, semanticJobs: this.semanticJobs };
-  }
-
   recordUsage(tokens: number, messageEstimate: number, environmentHash: string): void {
     if (!Number.isFinite(tokens) || tokens <= 0) return;
     this.usageBaseline = { tokens, messageEstimate, environmentHash };
@@ -277,11 +85,27 @@ export class SegmentIndex {
     environmentHash: string;
     environmentTokens: number;
     contextWindow: number;
+    maskedResults: number;
+    reclaimedTokens: number;
   }): ProjectionSnapshot {
     const sameTail = messages.length === this.sourceMessages.length
       && messages.length > 0
       && messageKey(messages[messages.length - 1]) === this.sourceKeys[this.sourceKeys.length - 1];
-    if (this.cached && sameTail && this.cached.environmentHash === params.environmentHash) return this.cached;
+    if (this.cached && sameTail && this.cached.environmentHash === params.environmentHash) {
+      if (this.cached.maskedResults === params.maskedResults
+        && this.cached.reclaimedTokens === params.reclaimedTokens) {
+        return this.cached;
+      }
+      // Masking stats changed but the message list did not: publish a new
+      // immutable snapshot object instead of re-projecting or mutating.
+      this.cached = {
+        ...this.cached,
+        maskedResults: params.maskedResults,
+        reclaimedTokens: params.reclaimedTokens,
+        rawTokens: this.cached.effectiveTokens + params.reclaimedTokens,
+      };
+      return this.cached;
+    }
 
     const appendOnly = this.sourceMessages.length > 0
       && messages.length >= this.sourceMessages.length
@@ -301,17 +125,14 @@ export class SegmentIndex {
 
     for (let index = start; index < messages.length; index++) {
       const message = messages[index];
-      const projected = this.projectMessage(message);
       this.sourceMessages.push(message);
       this.sourceKeys.push(messageKey(message));
-      this.projectedMessages.push(projected);
+      this.projectedMessages.push(message);
       this.rawEstimates.push(estimateTokens(message as any));
-      this.effectiveEstimates.push(estimateTokens(projected as any));
+      this.effectiveEstimates.push(estimateTokens(message as any));
     }
 
-    const rawMessageTokens = this.rawEstimates.reduce((sum, value) => sum + value, 0);
     const effectiveMessageTokens = this.effectiveEstimates.reduce((sum, value) => sum + value, 0);
-    const rawTokens = rawMessageTokens + params.environmentTokens;
     const baseline = this.usageBaseline;
     const usageBacked = baseline?.environmentHash === params.environmentHash;
     const effectiveTokens = usageBacked
@@ -329,55 +150,19 @@ export class SegmentIndex {
       environmentHash: params.environmentHash,
       messages: this.projectedMessages,
       messageEstimate: effectiveMessageTokens,
-      rawTokens,
+      rawTokens: effectiveTokens + params.reclaimedTokens,
       effectiveTokens,
-      reclaimedTokens: Math.max(0, rawTokens - effectiveTokens),
+      reclaimedTokens: params.reclaimedTokens,
       contextWindow: params.contextWindow,
       confidence,
       lowerBound,
       upperBound,
       segmentCount: this.sourceMessages.length,
-      semanticJobs: this.semanticJobs,
+      maskedResults: params.maskedResults,
     };
     return this.cached;
   }
 
-  private projectMessage(message: any): any {
-    if (message?.role !== "toolResult" || typeof message.toolCallId !== "string") return message;
-    const record = this.records.get(sourceKey(message.toolCallId, message.timestamp));
-    if (!record) return message;
-    const original = contentText(message.content);
-    const summary = record.semanticSummary ?? record.deterministicSummary;
-    const projected = `${summary}\nSource: ${record.sourceKey}; recover with context_tree_query.`;
-    if (projected.length >= original.length) return message;
-    return {
-      ...message,
-      content: [{ type: "text", text: projected }],
-    };
-  }
-
-  query(ids: string[]): ToolHistoryRecord[] {
-    const found: ToolHistoryRecord[] = [];
-    const seen = new Set<string>();
-    for (const id of ids) {
-      const direct = this.records.get(id);
-      const keys = direct ? [id] : this.ids.get(id) ?? [];
-      for (const key of keys) {
-        if (seen.has(key)) continue;
-        const record = this.records.get(key);
-        if (record) {
-          found.push(record);
-          seen.add(key);
-        }
-      }
-    }
-    return found;
-  }
-
-  allRecords(): ToolHistoryRecord[] {
-    return [...this.records.values()];
-  }
-
   getSnapshot(): ProjectionSnapshot | undefined {
     return this.cached;
   }

pi-compaction/src/semantic-worker.ts

@@ -1,189 +0,0 @@
-import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
-import type { PiCompactionConfig, ToolHistoryRecord } from "./types.js";
-
-interface Job {
-  sessionId: string;
-  sourceKey: string;
-  sourceHash: string;
-  toolName: string;
-  args: unknown;
-  isError: boolean;
-  rawText: string;
-  ctx: ExtensionContext;
-  config: PiCompactionConfig;
-  controller: AbortController;
-}
-
-export interface SemanticResult {
-  sessionId: string;
-  sourceKey: string;
-  sourceHash: string;
-  summary: string;
-}
-
-function modelFor(config: PiCompactionConfig, ctx: ExtensionContext): any {
-  if (config.semantic.model === "default") return ctx.model;
-  const slash = config.semantic.model.indexOf("/");
-  if (slash <= 0) return ctx.model;
-  return ctx.modelRegistry.find(
-    config.semantic.model.slice(0, slash),
-    config.semantic.model.slice(slash + 1),
-  ) ?? ctx.model;
-}
-
-function serialize(job: Job): string {
-  const args = (() => {
-    try { return JSON.stringify(job.args); } catch { return "[unserializable]"; }
-  })();
-  const max = job.config.semantic.maxInputChars;
-  const raw = job.rawText.length <= max
-    ? job.rawText
-    : `${job.rawText.slice(0, Math.floor(max * 0.7))}\n… omitted …\n${job.rawText.slice(-Math.floor(max * 0.3))}`;
-  return `Summarize this completed coding-agent tool call for future context.\nKeep concrete outcomes, paths, identifiers, values, and exact errors.\nUse at most 6 concise bullets. Do not give advice and do not call tools.\n\nTool: ${job.toolName}\nStatus: ${job.isError ? "ERROR" : "OK"}\nArgs: ${args}\nOutput:\n${raw}`;
-}
-
-async function run(job: Job): Promise<string> {
-  const model = modelFor(job.config, job.ctx);
-  if (!model) throw new Error("No summarizer model selected");
-  const message = await job.ctx.modelRegistry.complete(model, {
-    messages: [{
-      role: "user",
-      content: [{ type: "text", text: serialize(job) }],
-      timestamp: Date.now(),
-    }],
-  } as any, {
-    signal: job.controller.signal,
-    reasoningEffort: job.config.semantic.thinking === "off" ? undefined : job.config.semantic.thinking,
-  } as any);
-  if (message.stopReason === "length" || message.stopReason === "error" || message.stopReason === "aborted") {
-    throw new Error(message.errorMessage ?? `Semantic compaction stopped: ${message.stopReason}`);
-  }
-  const text = message.content
-    .flatMap((part: any) => part.type === "text" && typeof part.text === "string" ? [part.text] : [])
-    .join("\n")
-    .trim();
-  if (!text) throw new Error("Semantic compaction returned an empty summary");
-  return text;
-}
-
-function abortableDelay(ms: number, signal: AbortSignal): Promise<void> {
-  if (signal.aborted) return Promise.reject(new Error("Semantic compaction aborted"));
-  return new Promise((resolve, reject) => {
-    const timer = setTimeout(done, ms);
-    function done(): void {
-      signal.removeEventListener("abort", aborted);
-      resolve();
-    }
-    function aborted(): void {
-      clearTimeout(timer);
-      signal.removeEventListener("abort", aborted);
-      reject(new Error("Semantic compaction aborted"));
-    }
-    signal.addEventListener("abort", aborted, { once: true });
-  });
-}
-
-export class SemanticWorker {
-  private readonly queue: Job[] = [];
-  private readonly active = new Map<string, Job>();
-  private readonly ready: SemanticResult[] = [];
-  private readonly queuedKeys = new Set<string>();
-
-  enqueue(sessionId: string, record: ToolHistoryRecord, ctx: ExtensionContext, config: PiCompactionConfig): void {
-    if (!config.semantic.enabled || record.semanticSummary || record.rawText.length < config.semantic.minInputChars) return;
-    const key = `${sessionId}:${record.sourceKey}:${record.sourceHash}`;
-    if (this.queuedKeys.has(key)) return;
-    this.queuedKeys.add(key);
-    this.queue.push({
-      sessionId,
-      sourceKey: record.sourceKey,
-      sourceHash: record.sourceHash,
-      toolName: record.toolName,
-      args: record.args,
-      isError: record.isError,
-      rawText: record.rawText,
-      ctx,
-      config,
-      controller: new AbortController(),
-    });
-    this.pump(config.semantic.concurrency);
-  }
-
-  private pump(limit: number): void {
-    while (this.active.size < limit && this.queue.length > 0) {
-      const job = this.queue.shift()!;
-      const key = `${job.sessionId}:${job.sourceKey}:${job.sourceHash}`;
-      this.active.set(key, job);
-      void this.execute(key, job, limit);
-    }
-  }
-
-  private async execute(key: string, job: Job, limit: number): Promise<void> {
-    let lastError: unknown;
-    try {
-      for (let attempt = 0; attempt <= job.config.semantic.maxRetries; attempt++) {
-        try {
-          const summary = await run(job);
-          if (!job.controller.signal.aborted) {
-            this.ready.push({
-              sessionId: job.sessionId,
-              sourceKey: job.sourceKey,
-              sourceHash: job.sourceHash,
-              summary,
-            });
-          }
-          return;
-        } catch (error) {
-          lastError = error;
-          if (job.controller.signal.aborted || attempt === job.config.semantic.maxRetries) break;
-          await abortableDelay(Math.min(8_000, 1_000 * 2 ** attempt), job.controller.signal);
-        }
-      }
-      if (!job.controller.signal.aborted) {
-        console.error(`pi-compaction: semantic summary failed for ${job.sourceKey}:`, lastError);
-      }
-    } finally {
-      this.active.delete(key);
-      this.queuedKeys.delete(key);
-      this.pump(limit);
-    }
-  }
-
-  drain(sessionId: string): SemanticResult[] {
-    const drained: SemanticResult[] = [];
-    for (let index = this.ready.length - 1; index >= 0; index--) {
-      if (this.ready[index]!.sessionId !== sessionId) continue;
-      drained.push(this.ready[index]!);
-      this.ready.splice(index, 1);
-    }
-    return drained.reverse();
-  }
-
-  count(sessionId: string): number {
-    return this.queue.filter((job) => job.sessionId === sessionId).length
-      + [...this.active.values()].filter((job) => job.sessionId === sessionId).length;
-  }
-
-  cancelSession(sessionId: string): void {
-    for (const job of this.active.values()) {
-      if (job.sessionId === sessionId) job.controller.abort();
-    }
-    for (let index = this.queue.length - 1; index >= 0; index--) {
-      if (this.queue[index]!.sessionId === sessionId) {
-        const job = this.queue[index]!;
-        this.queuedKeys.delete(`${job.sessionId}:${job.sourceKey}:${job.sourceHash}`);
-        this.queue.splice(index, 1);
-      }
-    }
-    for (let index = this.ready.length - 1; index >= 0; index--) {
-      if (this.ready[index]!.sessionId === sessionId) this.ready.splice(index, 1);
-    }
-  }
-
-  cancelAll(): void {
-    for (const job of this.active.values()) job.controller.abort();
-    this.queue.length = 0;
-    this.ready.length = 0;
-    this.queuedKeys.clear();
-  }
-}

pi-compaction/src/service.ts

@@ -2,8 +2,8 @@ import {
   SERVICE_SYMBOL,
   type CompactionMetrics,
   type CompactionService,
+  type MaskedResultRecord,
   type ProjectionSnapshot,
-  type ToolHistoryRecord,
 } from "./types.js";
 
 interface GlobalWithCompactionService {
@@ -14,11 +14,11 @@ const root = globalThis as GlobalWithCompactionService;
 
 export class CompactionServiceRegistry implements CompactionService {
   private readonly snapshots = new Map<string, ProjectionSnapshot>();
-  private readonly queryHandlers = new Map<string, (ids: string[]) => ToolHistoryRecord[]>();
+  private readonly queryHandlers = new Map<string, (ids: string[]) => MaskedResultRecord[]>();
 
   setSession(
     sessionId: string,
-    query: (ids: string[]) => ToolHistoryRecord[],
+    query: (ids: string[]) => MaskedResultRecord[],
   ): void {
     this.queryHandlers.set(sessionId, query);
   }
@@ -58,11 +58,11 @@ export class CompactionServiceRegistry implements CompactionService {
       percent: snapshot.contextWindow > 0 ? 100 * snapshot.effectiveTokens / snapshot.contextWindow : 0,
       confidence: snapshot.confidence,
       segmentCount: snapshot.segmentCount,
-      semanticJobs: snapshot.semanticJobs,
+      maskedResults: snapshot.maskedResults,
     };
   }
 
-  query(sessionId: string, ids: string[]): ToolHistoryRecord[] {
+  query(sessionId: string, ids: string[]): MaskedResultRecord[] {
     return this.queryHandlers.get(sessionId)?.(ids) ?? [];
   }
 }

pi-compaction/src/token-accounting.ts

@@ -1,13 +1,5 @@
-import type { PiCompactionConfig, ProjectionSnapshot } from "./types.js";
+// Effective trigger for one configured (tokens, ratio) pair: the absolute
-
+// token figure wins on large windows, the ratio caps it on small ones.
-export type CompactionReason = "manual" | "threshold" | "overflow";
+export function effectiveThreshold(tokens: number, ratio: number, contextWindow: number): number {
-
+  return contextWindow > 0 ? Math.min(tokens, Math.floor(ratio * contextWindow)) : tokens;
-export function shouldCompact(
-  reason: CompactionReason,
-  snapshot: Pick<ProjectionSnapshot, "lowerBound" | "contextWindow">,
-  config: Pick<PiCompactionConfig, "hardRatio">,
-): boolean {
-  if (reason === "manual" || reason === "overflow") return true;
-  if (snapshot.contextWindow <= 0) return true;
-  return snapshot.lowerBound / snapshot.contextWindow >= config.hardRatio;
 }

pi-compaction/src/types.ts

@@ -5,58 +5,33 @@ export const COMPACTION_VERSION = 1 as const;
 export const QUERY_TOOL_NAME = "context_tree_query";
 export const SERVICE_SYMBOL = Symbol.for("my-pi.pi-compaction.service.v1");
 
+export interface MaskingConfig {
+  enabled: boolean;
+  triggerTokens: number;
+  triggerRatio: number;
+  targetTokens: number;
+  targetRatio: number;
+  minClearTokens: number;
+  keepRecentTurns: number;
+  minEntryTokens: number;
+  excludeTools: string[];
+}
+
+export interface SummaryConfig {
+  triggerTokens: number;
+  triggerRatio: number;
+  thinking: "off" | "minimal" | "low" | "medium" | "high" | "xhigh";
+}
+
 export interface PiCompactionConfig {
   enabled: boolean;
-  softRatio: number;
+  masking: MaskingConfig;
-  hardRatio: number;
+  summary: SummaryConfig;
-  targetRatio: number;
-  deterministic: {
-    enabled: boolean;
-    maxStubTokens: number;
-  };
-  semantic: {
-    enabled: boolean;
-    model: string;
-    thinking: "off" | "minimal" | "low" | "medium" | "high" | "xhigh";
-    concurrency: number;
-    minInputChars: number;
-    maxInputChars: number;
-    maxRetries: number;
-  };
   native: {
     enabled: boolean;
   };
 }
 
-export interface SegmentSummaryRecord {
-  version: typeof COMPACTION_VERSION;
-  kind: "segment-summary";
-  sourceKey: string;
-  toolCallId: string;
-  resultTimestamp: number;
-  toolName: string;
-  sourceHash: string;
-  deterministicSummary: string;
-  semanticSummary?: string;
-  rawChars: number;
-  effectiveChars: number;
-}
-
-export type PiCompactionRecord = SegmentSummaryRecord;
-
-export interface ToolHistoryRecord {
-  sourceKey: string;
-  toolCallId: string;
-  resultTimestamp: number;
-  toolName: string;
-  args: unknown;
-  isError: boolean;
-  rawText: string;
-  sourceHash: string;
-  deterministicSummary: string;
-  semanticSummary?: string;
-}
-
 export interface UsageBaseline {
   tokens: number;
   messageEstimate: number;
@@ -78,7 +53,7 @@ export interface ProjectionSnapshot {
   lowerBound: number;
   upperBound: number;
   segmentCount: number;
-  semanticJobs: number;
+  maskedResults: number;
   codexItems?: ResponseItem[];
 }
 
@@ -99,6 +74,13 @@ export interface PortableCheckpointDetails {
   sourceFingerprint: string;
 }
 
+/** What the service returns for one masked tool result — the same placeholder the model sees. */
+export interface MaskedResultRecord {
+  toolCallId: string;
+  resultTimestamp: number;
+  deterministicSummary: string;
+}
+
 export interface CompactionMetrics {
   sessionId: string;
   generation: number;
@@ -109,11 +91,11 @@ export interface CompactionMetrics {
   percent: number;
   confidence: ProjectionSnapshot["confidence"];
   segmentCount: number;
-  semanticJobs: number;
+  maskedResults: number;
 }
 
 export interface CompactionService {
   getSnapshot(sessionId: string): ProjectionSnapshot | undefined;
   getMetrics(sessionId: string): CompactionMetrics | undefined;
-  query(sessionId: string, ids: string[]): ToolHistoryRecord[];
+  query(sessionId: string, ids: string[]): MaskedResultRecord[];
 }

pi-compaction/test/config.test.ts

@@ -2,34 +2,82 @@ import test from "node:test";
 import assert from "node:assert/strict";
 import { DEFAULT_CONFIG, normalizeConfig } from "../src/config.ts";
 
-test("normalizes one greenfield piCompaction config without legacy aliases", () => {
+test("normalizes a full pi-compaction config", () => {
   const config = normalizeConfig({
     enabled: false,
-    softRatio: 0.7,
+    masking: {
-    hardRatio: 0.91,
+      enabled: false,
-    targetRatio: 0.5,
+      triggerTokens: 150_000,
-    deterministic: { enabled: false, maxStubTokens: 512 },
+      triggerRatio: 0.5,
-    semantic: { enabled: false, concurrency: 2, maxRetries: 0 },
+      targetTokens: 60_000,
+      targetRatio: 0.2,
+      minClearTokens: 30_000,
+      keepRecentTurns: 5,
+      minEntryTokens: 512,
+      excludeTools: ["bash", "read"],
+    },
+    summary: { triggerTokens: 300_000, triggerRatio: 0.8, thinking: "medium" },
     native: { enabled: false },
   });
   assert.equal(config.enabled, false);
-  assert.equal(config.deterministic.enabled, false);
+  assert.deepEqual(config.masking, {
-  assert.equal(config.deterministic.maxStubTokens, 512);
+    enabled: false,
-  assert.equal(config.semantic.enabled, false);
+    triggerTokens: 150_000,
-  assert.equal(config.semantic.maxRetries, 0);
+    triggerRatio: 0.5,
+    targetTokens: 60_000,
+    targetRatio: 0.2,
+    minClearTokens: 30_000,
+    keepRecentTurns: 5,
+    minEntryTokens: 512,
+    excludeTools: ["bash", "read"],
+  });
+  assert.equal(config.summary.triggerTokens, 300_000);
+  assert.equal(config.summary.triggerRatio, 0.8);
+  assert.equal(config.summary.thinking, "medium");
   assert.equal(config.native.enabled, false);
 });
 
-test("invalid ratios and bounds fail safe to bounded defaults", () => {
+test("invalid fields fail safe to bounded defaults", () => {
   const config = normalizeConfig({
-    softRatio: 2,
+    enabled: "yes",
-    hardRatio: -1,
+    masking: {
-    targetRatio: 0.99,
+      triggerTokens: -1,
-    semantic: { concurrency: 50, maxRetries: -1 },
+      triggerRatio: 2,
+      targetTokens: "a lot",
+      targetRatio: 0,
+      minClearTokens: 0,
+      keepRecentTurns: 500,
+      minEntryTokens: -3,
+      excludeTools: "read",
+    },
+    summary: { triggerTokens: Number.NaN, triggerRatio: 1.5, thinking: "insane" },
+    native: { enabled: 1 },
   });
-  assert.equal(config.softRatio, DEFAULT_CONFIG.softRatio);
+  assert.equal(config.enabled, DEFAULT_CONFIG.enabled);
-  assert.ok(config.targetRatio < config.softRatio);
+  assert.deepEqual(config.masking, DEFAULT_CONFIG.masking);
-  assert.ok(config.hardRatio > config.softRatio);
+  assert.equal(config.summary.triggerTokens, DEFAULT_CONFIG.summary.triggerTokens);
-  assert.equal(config.semantic.concurrency, DEFAULT_CONFIG.semantic.concurrency);
+  assert.equal(config.summary.triggerRatio, DEFAULT_CONFIG.summary.triggerRatio);
-  assert.equal(config.semantic.maxRetries, DEFAULT_CONFIG.semantic.maxRetries);
+  assert.equal(config.summary.thinking, DEFAULT_CONFIG.summary.thinking);
+  assert.equal(config.native.enabled, DEFAULT_CONFIG.native.enabled);
+});
+
+test("masking target is corrected below trigger and summary above masking", () => {
+  const config = normalizeConfig({
+    masking: { triggerTokens: 80_000, triggerRatio: 0.5, targetTokens: 120_000, targetRatio: 0.9 },
+    summary: { triggerTokens: 50_000, triggerRatio: 0.4 },
+  });
+  assert.equal(config.masking.targetTokens, 40_000);
+  assert.equal(config.masking.targetRatio, 0.25);
+  assert.equal(config.summary.triggerTokens, 160_000);
+  assert.ok(config.summary.triggerRatio > config.masking.triggerRatio);
+});
+
+test("minClearTokens and excludeTools accept valid overrides", () => {
+  const config = normalizeConfig({
+    masking: { minClearTokens: 1, keepRecentTurns: 1, minEntryTokens: 0, excludeTools: [] },
+  });
+  assert.equal(config.masking.minClearTokens, 1);
+  assert.equal(config.masking.keepRecentTurns, 1);
+  assert.equal(config.masking.minEntryTokens, 0);
+  assert.deepEqual(config.masking.excludeTools, []);
 });

pi-compaction/test/extension-runtime.test.ts

@@ -1,7 +1,8 @@
 import test from "node:test";
 import assert from "node:assert/strict";
-import { performance } from "node:perf_hooks";
 import piCompaction from "../index.ts";
+import { getCompactionService } from "../src/service.ts";
+import { MASK_PREFIX } from "../src/masking.ts";
 
 function harness() {
   const handlers = new Map<string, Function[]>();
@@ -24,90 +25,253 @@ function harness() {
   return { pi, handlers, tools, commands, entries };
 }
 
-function context(branch: any[] = []) {
+function context(branch: any[] = [], tokens = 1_000) {
-  return {
+  const compactCalls: any[] = [];
+  const sessionId = `session-${Math.random().toString(36).slice(2)}`;
+  const ctx = {
     sessionManager: {
-      getSessionId: () => "session-test",
+      getSessionId: () => sessionId,
       getBranch: () => branch,
+      getLeafId: () => "leaf-1",
     },
     model: { provider: "test", api: "test", id: "model", contextWindow: 100_000 },
     modelRegistry: {},
     getSystemPrompt: () => "system",
-    getContextUsage: () => ({ tokens: 1_000, contextWindow: 100_000, percent: 1 }),
+    getContextUsage: () => ({ tokens, contextWindow: 100_000, percent: 1 }),
+    compact(options: unknown) { compactCalls.push(options); },
+    isIdle: () => true,
     hasUI: false,
     abort() {},
   } as any;
+  return { ctx, compactCalls };
 }
 
-test("one runtime projects tool evidence and recovers it through context_tree_query", async () => {
+function projectedEntry(id: string, message: any) {
+  return { sourceEntry: { id, type: "message", message }, messages: [message] };
+}
+
+function assistantMessage(id: string, calls: Array<{ id: string; name: string; args?: unknown }>) {
+  return {
+    role: "assistant",
+    content: calls.map((call) => ({ type: "toolCall", id: call.id, name: call.name, arguments: call.args ?? {} })),
+    timestamp: id,
+  };
+}
+
+function toolResultMessage(id: string, toolCallId: string, text: string, toolName = "bash") {
+  return {
+    role: "toolResult",
+    toolCallId,
+    toolName,
+    content: [{ type: "text", text }],
+    timestamp: id,
+  };
+}
+
+// Default config: masking trigger = min(200K, 60% of window); at the harness
+// 100K window that is 60K tokens and the target is 35K, so 60K used needs
+// ~25K tokens of maskable results outside the 10-turn protected window.
+function pressurizedContextEntries() {
+  const entries = [
+    projectedEntry("a1", assistantMessage("a1", [{ id: "c1", name: "bash", args: { command: "npm test" } }])),
+    projectedEntry("r1", toolResultMessage("r1", "c1", `failure evidence\n${"x".repeat(120_000)}\nstack tail`)),
+  ];
+  for (let index = 0; index < 10; index++) {
+    entries.push(projectedEntry(`keep-${index}`, assistantMessage(`keep-${index}`, [])));
+  }
+  return entries;
+}
+
+test("turn_end masks old tool results under pressure and preserves prior drafts", async () => {
   const h = harness();
   piCompaction(h.pi);
-  const ctx = context();
+  const { ctx } = context([], 60_000);
   await h.handlers.get("session_start")![0]!({}, ctx);
 
-  const assistant = {
+  const prior = { type: "custom", customType: "other-extension", data: {} };
-    role: "assistant",
+  const result = await h.handlers.get("turn_end")![0]!({
-    content: [{ type: "toolCall", id: "call-1", name: "bash", arguments: { command: "npm test" } }],
+    entries: [prior],
-    timestamp: 10,
+    message: assistantMessage("a1", []),
-  };
+    toolResults: [],
-  const result = {
+    context: { contextEntries: pressurizedContextEntries(), contextMessages: [] },
-    role: "toolResult",
+  }, ctx);
-    toolCallId: "call-1",
-    toolName: "bash",
-    content: [{ type: "text", text: `failure evidence\n${"x".repeat(12_000)}\nstack tail` }],
-    isError: true,
-    timestamp: 11,
-  };
-  await h.handlers.get("turn_end")![0]!({ message: assistant, toolResults: [result] }, ctx);
-  const projected = await h.handlers.get("context")![0]!({ messages: [assistant, result] }, ctx);
 
-  assert.match(projected.messages[1].content[0].text, /pi-compaction tool=bash/);
+  assert.ok(result?.entries);
-  const cacheStart = performance.now();
+  assert.strictEqual(result.entries[0], prior);
-  for (let iteration = 0; iteration < 1_000; iteration++) {
+  const edits = result.entries.slice(1);
-    await h.handlers.get("context")![0]!({ messages: [assistant, result] }, ctx);
+  assert.equal(edits.length, 1);
-  }
+  assert.equal(edits[0]!.type, "context_edit");
-  const cacheAverageMs = (performance.now() - cacheStart) / 1_000;
+  assert.equal(edits[0]!.targetId, "r1");
-  assert.ok(cacheAverageMs < 2, `full context handler cache-hit average ${cacheAverageMs.toFixed(3)}ms exceeded 2ms`);
+  assert.ok(edits[0]!.replacement.content.startsWith(MASK_PREFIX));
-  assert.equal(h.entries.length, 1);
+  // Masking is append-only context_edit; nothing is written as custom entries.
-  assert.equal(h.entries[0]!.customType, "pi-compaction");
+  assert.equal(h.entries.length, 0);
-
-  const query = h.tools.get("context_tree_query");
-  const recovered = await query.execute("q", { toolCallIds: ["call-1"] }, new AbortController().signal, undefined, ctx);
-  assert.match(recovered.content[0].text, /failure evidence/);
 
   await h.handlers.get("session_shutdown")![0]!({}, ctx);
 });
 
-test("small tool results are passed through without a stub wrapper", async () => {
+test("turn_end below the masking trigger returns undefined", async () => {
   const h = harness();
   piCompaction(h.pi);
-  const ctx = context();
+  const { ctx } = context([], 40_000);
   await h.handlers.get("session_start")![0]!({}, ctx);
 
-  const assistant = {
+  const result = await h.handlers.get("turn_end")![0]!({
-    role: "assistant",
+    entries: [],
-    content: [{ type: "toolCall", id: "call-small", name: "bash", arguments: { command: "ls" } }],
+    message: assistantMessage("a1", []),
-    timestamp: 20,
+    toolResults: [],
-  };
+    context: { contextEntries: pressurizedContextEntries(), contextMessages: [] },
-  const result = {
+  }, ctx);
-    role: "toolResult",
+  assert.equal(result, undefined);
-    toolCallId: "call-small",
+  await h.handlers.get("session_shutdown")![0]!({}, ctx);
-    toolName: "bash",
+});
-    content: [{ type: "text", text: "ok\nfile-a.ts\nfile-b.ts" }],
-    timestamp: 21,
-  };
 
-  await h.handlers.get("turn_end")![0]!({ message: assistant, toolResults: [result] }, ctx);
+test("context hook returns messages untouched on the non-checkpoint path", async () => {
-  const projected = await h.handlers.get("context")![0]!({ messages: [assistant, result] }, ctx);
+  const h = harness();
+  piCompaction(h.pi);
+  const { ctx } = context();
+  await h.handlers.get("session_start")![0]!({}, ctx);
 
-  assert.strictEqual(projected.messages[1].content[0].text, "ok\nfile-a.ts\nfile-b.ts");
+  const messages = [assistantMessage("a1", []), toolResultMessage("r1", "c1", "visible\nfile-a.ts")];
-  assert.doesNotMatch(projected.messages[1].content[0].text, /pi-compaction tool=/);
+  const projected = await h.handlers.get("context")![0]!({ messages }, ctx);
+  assert.equal(projected, undefined);
+  await h.handlers.get("session_shutdown")![0]!({}, ctx);
+});
+
+test("agent_settled compacts once above the summary threshold and never repeats on the same leaf", async () => {
+  const h = harness();
+  piCompaction(h.pi);
+  const { ctx, compactCalls } = context([], 95_000);
+  await h.handlers.get("session_start")![0]!({}, ctx);
+
+  h.handlers.get("agent_settled")![0]!({ type: "agent_settled", aborted: false }, ctx);
+  assert.equal(compactCalls.length, 1);
+  assert.equal(typeof compactCalls[0].onError, "function");
+
+  h.handlers.get("agent_settled")![0]!({ type: "agent_settled", aborted: false }, ctx);
+  assert.equal(compactCalls.length, 1);
+
+  ctx.sessionManager.getLeafId = () => "leaf-2";
+  h.handlers.get("agent_settled")![0]!({ type: "agent_settled", aborted: false }, ctx);
+  assert.equal(compactCalls.length, 2);
+  await h.handlers.get("session_shutdown")![0]!({}, ctx);
+});
+
+test("agent_settled stays idle below the summary threshold", async () => {
+  const h = harness();
+  piCompaction(h.pi);
+  const { ctx, compactCalls } = context([], 50_000);
+  await h.handlers.get("session_start")![0]!({}, ctx);
+
+  h.handlers.get("agent_settled")![0]!({ type: "agent_settled", aborted: false }, ctx);
+  assert.equal(compactCalls.length, 0);
+  await h.handlers.get("session_shutdown")![0]!({}, ctx);
+});
+
+test("context_tree_query recovers the original tool output from the branch", async () => {
+  const h = harness();
+  piCompaction(h.pi);
+  const branch = [
+    { type: "message", id: "a1", parentId: null, timestamp: "t", message: assistantMessage("a1", [{ id: "c1", name: "bash", args: { command: "npm test" } }]) },
+    { type: "message", id: "r1", parentId: "a1", timestamp: "t", message: toolResultMessage("r1", "c1", "failure evidence\nstack tail") },
+  ];
+  const { ctx } = context(branch);
+  await h.handlers.get("session_start")![0]!({}, ctx);
 
-  // Original evidence stays recoverable even when the message is passed through.
   const query = h.tools.get("context_tree_query")!;
-  const recovered = await query.execute("q", { toolCallIds: ["call-small"] }, new AbortController().signal, undefined, ctx);
+  const recovered = await query.execute("q", { toolCallIds: ["c1"] }, new AbortController().signal, undefined, ctx);
-  assert.match(recovered.content[0].text, /file-a\.ts/);
+  assert.match(recovered.content[0].text, /failure evidence/);
+  assert.match(recovered.content[0].text, /stack tail/);
+  assert.deepEqual(recovered.details.missing, []);
 
+  const missing = await query.execute("q", { toolCallIds: ["nope"] }, new AbortController().signal, undefined, ctx);
+  assert.match(missing.content[0].text, /not found/);
+  await h.handlers.get("session_shutdown")![0]!({}, ctx);
+});
+
+test("service query returns placeholders only for masked tool call ids", async () => {
+  const h = harness();
+  piCompaction(h.pi);
+  const { ctx } = context([], 60_000);
+  await h.handlers.get("session_start")![0]!({}, ctx);
+  const sessionId = ctx.sessionManager.getSessionId();
+
+  await h.handlers.get("turn_end")![0]!({
+    entries: [],
+    message: assistantMessage("a1", []),
+    toolResults: [],
+    context: { contextEntries: pressurizedContextEntries(), contextMessages: [] },
+  }, ctx);
+
+  const records = getCompactionService()!.query(sessionId, ["c1", "never-masked"]);
+  assert.equal(records.length, 1);
+  assert.equal(records[0]!.toolCallId, "c1");
+  assert.ok(records[0]!.deterministicSummary.startsWith(MASK_PREFIX));
+  assert.equal(typeof records[0]!.resultTimestamp, "number");
+  await h.handlers.get("session_shutdown")![0]!({}, ctx);
+});
+
+test("hydrating a branch with this extension's context_edit rebuilds masking metrics", async () => {
+  const h = harness();
+  piCompaction(h.pi);
+  const placeholder = `${MASK_PREFIX}bash result cleared to save context (~1_000 tokens). Call args: {}. Re-run the call if still needed, or recover the original with context_tree_query {"toolCallIds":["c1"]}.`;
+  const branch = [
+    { type: "message", id: "a1", parentId: null, timestamp: "t", message: assistantMessage("a1", [{ id: "c1", name: "bash", args: { command: "npm test" } }]) },
+    { type: "message", id: "r1", parentId: "a1", timestamp: "t", message: toolResultMessage("r1", "c1", "x".repeat(8_000)) },
+    { type: "context_edit", id: "e1", parentId: "r1", timestamp: "t", targetId: "r1", replacement: { content: placeholder } },
+  ];
+  const { ctx } = context(branch);
+  await h.handlers.get("session_start")![0]!({}, ctx);
+  const sessionId = ctx.sessionManager.getSessionId();
+
+  await h.handlers.get("context")![0]!({ messages: [toolResultMessage("r1", "c1", "x".repeat(8_000))] }, ctx);
+  const metrics = getCompactionService()!.getMetrics(sessionId);
+  assert.equal(metrics!.maskedResults, 1);
+  assert.ok(metrics!.reclaimedTokens > 0);
+  assert.ok(metrics!.rawTokens > metrics!.effectiveTokens);
+  await h.handlers.get("session_shutdown")![0]!({}, ctx);
+});
+
+test("session_before_compact no longer cancels threshold compactions", async () => {
+  const h = harness();
+  piCompaction(h.pi);
+  const { ctx } = context();
+  await h.handlers.get("session_start")![0]!({}, ctx);
+  ctx.modelRegistry.getApiKeyAndHeaders = async () => ({ ok: false, error: "no credentials" });
+
+  const result = await h.handlers.get("session_before_compact")![0]!({
+    reason: "threshold", branchEntries: [], preparation: { firstKeptEntryId: "u1" },
+    signal: new AbortController().signal,
+  }, ctx);
+  assert.equal(result, undefined);
+  await h.handlers.get("session_shutdown")![0]!({}, ctx);
+});
+
+test("native checkpoint projection applies context edits made after the checkpoint", async () => {
+  const h = harness();
+  piCompaction(h.pi);
+  const details = {
+    kind: "pi-compaction",
+    version: 1,
+    checkpoint: "native",
+    modelKey: "test:test:model",
+    replacementHistory: [{ type: "compaction", encrypted_content: "opaque" }],
+    portableSummary: "checkpoint summary",
+    sourceFingerprint: "fp",
+  };
+  const branch = [
+    { type: "compaction", id: "c1", parentId: null, timestamp: "t", summary: "marker", firstKeptEntryId: "r1", tokensBefore: 1_000, details },
+    { type: "message", id: "r1", parentId: "c1", timestamp: "t", message: toolResultMessage("r1", "c1", "ORIGINAL-CONTENT-HERE") },
+    { type: "context_edit", id: "e1", parentId: "r1", timestamp: "t", targetId: "r1", replacement: { content: `${MASK_PREFIX}bash result cleared to save context (~5 tokens). Call args: {}.` } },
+    { type: "message", id: "u1", parentId: "e1", timestamp: "t", message: { role: "user", content: "next question", timestamp: 30 } },
+  ] as any[];
+  const { ctx } = context(branch);
+  await h.handlers.get("session_start")![0]!({}, ctx);
+
+  const projected = await h.handlers.get("context")![0]!({ messages: [] }, ctx);
+  const serialized = JSON.stringify(projected.messages);
+  assert.doesNotMatch(serialized, /ORIGINAL-CONTENT-HERE/);
+  assert.match(serialized, /result cleared to save context/);
+  assert.match(serialized, /next question/);
   await h.handlers.get("session_shutdown")![0]!({}, ctx);
 });
 
@@ -127,7 +291,7 @@ test("model switching after a native checkpoint projects the portable checkpoint
     { type: "compaction", id: "c1", parentId: null, timestamp: new Date().toISOString(), summary: "marker", firstKeptEntryId: "m0", tokensBefore: 1_000, details },
     { type: "message", id: "m1", parentId: "c1", timestamp: new Date().toISOString(), message: { role: "user", content: "continue", timestamp: 20 } },
   ] as any[];
-  const ctx = context(branch);
+  const { ctx } = context(branch);
   await h.handlers.get("session_start")![0]!({}, ctx);
   const projected = await h.handlers.get("context")![0]!({ messages: [] }, ctx);
   assert.match(projected.messages[0].content[0].text, /decisions and files/);
@@ -143,7 +307,7 @@ for (const provider of ["sub-api", "openai-codex"]) {
       type: "message", id: "u1", parentId: null,
       message: { role: "user", content: "Remember the test decision.", timestamp: 1 },
     }];
-    const ctx = context(branch);
+    const { ctx } = context(branch);
     ctx.model = {
       provider, api: provider === "sub-api" ? "openai-responses" : "openai-codex-responses",
       id: "selected-model", input: ["text"], contextWindow: 100_000,
@@ -231,7 +395,7 @@ test("SubAPI native failures and cancellation return control to Pi without a che
     type: "message", id: "u1", parentId: null,
     message: { role: "user", content: "Test only.", timestamp: 1 },
   }];
-  const ctx = context(branch);
+  const { ctx } = context(branch);
   ctx.model = { provider: "sub-api", api: "openai-responses", id: "test-model", input: ["text"], baseUrl: "https://gateway.example/v1" };
   ctx.modelRegistry.getApiKeyAndHeaders = async () => ({ ok: true, apiKey: "test-key" });
   const responses = [
@@ -266,7 +430,7 @@ test("SubAPI native failures and cancellation return control to Pi without a che
 test("other providers and SubAPI Chat Completions keep their provider requests unchanged", async () => {
   const h = harness();
   piCompaction(h.pi);
-  const ctx = context();
+  const { ctx } = context();
   try {
     for (const model of [
       { provider: "other", api: "openai-responses", id: "test" },
@@ -302,7 +466,7 @@ test("native runtime delegates authentication to each provider without blocking
         type: "message", id: "u1", parentId: null,
         message: { role: "user", content: "Synthetic authentication test.", timestamp: 1 },
       }];
-      const ctx = context(branch);
+      const { ctx } = context(branch);
       const provider = scenario.provider ?? "sub-api";
       ctx.model = {
         provider, api: provider === "sub-api" ? "openai-responses" : "openai-codex-responses",

pi-compaction/test/masking.test.ts

@@ -0,0 +1,209 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { MASK_PREFIX, planMasking, rebuildMaskedLedger } from "../src/masking.ts";
+import { effectiveThreshold } from "../src/token-accounting.ts";
+import type { MaskingConfig } from "../src/types.ts";
+
+const MASKING: MaskingConfig = {
+  enabled: true,
+  triggerTokens: 1_000,
+  triggerRatio: 0.9,
+  targetTokens: 400,
+  targetRatio: 0.35,
+  minClearTokens: 200,
+  keepRecentTurns: 2,
+  minEntryTokens: 10,
+  excludeTools: ["ask_user_question"],
+};
+
+const estimate = (message: any): number => Math.ceil(JSON.stringify(message).length / 4);
+
+function assistantMessage(id: string, calls: Array<{ id: string; name: string; args?: unknown }>) {
+  return {
+    role: "assistant",
+    content: calls.map((call) => ({ type: "toolCall", id: call.id, name: call.name, arguments: call.args ?? {} })),
+    timestamp: id,
+  };
+}
+
+function toolResultMessage(id: string, toolCallId: string, text: string, toolName = "bash") {
+  return {
+    role: "toolResult",
+    toolCallId,
+    toolName,
+    content: [{ type: "text", text }],
+    timestamp: id,
+  };
+}
+
+function entry(id: string, message: any, messages = [message]) {
+  return { sourceEntry: { id, type: "message", message }, messages };
+}
+
+function protect(turns = MASKING.keepRecentTurns) {
+  return Array.from({ length: turns }, (_unused, index) =>
+    entry(`recent-${index}`, { role: "assistant", content: [{ type: "text", text: "ok" }], timestamp: `recent-${index}` }));
+}
+
+test("below the effective trigger plans no drafts", () => {
+  const entries = [
+    entry("e1", assistantMessage("a1", [{ id: "c1", name: "bash", args: { command: "ls" } }])),
+    entry("e2", toolResultMessage("r1", "c1", "x".repeat(8_000))),
+    ...protect(),
+  ];
+  const plan = planMasking({ contextEntries: entries, currentTokens: 999, contextWindow: 0, masking: MASKING, estimate });
+  assert.equal(plan.drafts.length, 0);
+  assert.equal(plan.masked.length, 0);
+});
+
+test("above the trigger masks oldest first until context fits under target", () => {
+  const entries = [
+    entry("e1", assistantMessage("a1", [
+      { id: "c1", name: "bash", args: { command: "one" } },
+      { id: "c2", name: "bash", args: { command: "two" } },
+      { id: "c3", name: "bash", args: { command: "three" } },
+    ])),
+    entry("e2", toolResultMessage("r1", "c1", "a".repeat(4_000))),
+    entry("e3", toolResultMessage("r2", "c2", "b".repeat(4_000))),
+    entry("e4", toolResultMessage("r3", "c3", "c".repeat(4_000))),
+    entry("e5", toolResultMessage("r4", "c4", "recent result".repeat(200))),
+    ...protect(),
+  ];
+  // needed = 2_000 - 400 = 1_600; each result clears ~980 tokens.
+  const plan = planMasking({ contextEntries: entries, currentTokens: 2_000, contextWindow: 0, masking: MASKING, estimate });
+  assert.deepEqual(plan.drafts.map((draft) => draft.targetId), ["e2", "e3"]);
+});
+
+test("results inside the recent keepRecentTurns window are never masked", () => {
+  const entries = [
+    entry("e1", toolResultMessage("r1", "c1", "a".repeat(4_000))),
+    entry("recent-a", assistantMessage("ra", [{ id: "rc", name: "bash", args: {} }])),
+    entry("recent-r", toolResultMessage("rr", "rc", "b".repeat(4_000))),
+    entry("recent-b", assistantMessage("rb", [])),
+  ];
+  const plan = planMasking({ contextEntries: entries, currentTokens: 3_500, contextWindow: 0, masking: MASKING, estimate });
+  assert.deepEqual(plan.drafts.map((draft) => draft.targetId), ["e1"]);
+});
+
+test("excluded tools, SKILL.md reads, already-masked and small results are skipped", () => {
+  const entries = [
+    entry("e1", assistantMessage("a1", [
+      { id: "c1", name: "ask_user_question", args: { question: "x" } },
+      { id: "c2", name: "read", args: { path: "/repo/skills/foo/SKILL.md" } },
+      { id: "c3", name: "bash", args: { command: "old" } },
+      { id: "c4", name: "bash", args: { command: "tiny" } },
+    ])),
+    entry("e2", toolResultMessage("r1", "c1", "a".repeat(4_000), "ask_user_question")),
+    entry("e3", toolResultMessage("r2", "c2", "b".repeat(4_000), "read")),
+    entry("e4", toolResultMessage("r3", "c3", `${MASK_PREFIX}bash result cleared to save context (~900 tokens).`), ),
+    entry("e5", toolResultMessage("r4", "c4", "tiny")),
+    entry("e6", toolResultMessage("r5", "c5", "d".repeat(4_000))),
+    ...protect(),
+  ];
+  const masking = { ...MASKING, minClearTokens: 100, minEntryTokens: 100 };
+  const plan = planMasking({ contextEntries: entries, currentTokens: 3_500, contextWindow: 0, masking, estimate });
+  assert.deepEqual(plan.drafts.map((draft) => draft.targetId), ["e6"]);
+  assert.equal(plan.masked[0]!.toolCallId, "c5");
+});
+
+test("when clearable volume stays under minClear nothing is masked", () => {
+  const entries = [
+    entry("e1", toolResultMessage("r1", "c1", "a".repeat(600))),
+    ...protect(),
+  ];
+  const plan = planMasking({ contextEntries: entries, currentTokens: 3_500, contextWindow: 0, masking: MASKING, estimate });
+  // ~130 tokens clearable, below minClearTokens (200) → not worth invalidating cache.
+  assert.equal(plan.drafts.length, 0);
+});
+
+test("placeholder is deterministic, names the tool and call id, and truncates args at 300 chars", () => {
+  const longArgs = { command: `run ${"v".repeat(600)}` };
+  const entries = [
+    entry("e1", assistantMessage("a1", [{ id: "c1", name: "bash", args: longArgs }])),
+    entry("e2", toolResultMessage("r1", "c1", "a".repeat(4_000))),
+    ...protect(),
+  ];
+  const masking = { ...MASKING, minClearTokens: 1 };
+  const first = planMasking({ contextEntries: entries, currentTokens: 3_500, contextWindow: 0, masking, estimate });
+  const second = planMasking({ contextEntries: entries, currentTokens: 3_500, contextWindow: 0, masking, estimate });
+  assert.equal(first.drafts.length, 1);
+  const placeholder = first.masked[0]!.placeholder;
+  assert.equal(placeholder, second.masked[0]!.placeholder);
+  assert.ok(placeholder.startsWith(MASK_PREFIX));
+  assert.match(placeholder, /bash result cleared/);
+  assert.match(placeholder, /"toolCallIds":\["c1"\]/);
+  const argsText = placeholder.match(/Call args: (.*)\. Re-run/)?.[1] ?? "";
+  assert.ok(argsText.length <= 301, `args excerpt ${argsText.length} exceeded 300+ellipsis`);
+  assert.ok(placeholder.includes("…"));
+});
+
+test("a masked result carrying details.overflow.id gets an overflow_query hint in the placeholder", () => {
+  const withOverflow = {
+    ...toolResultMessage("r1", "c1", "a".repeat(4_000)),
+    details: { overflow: { id: "0123456789ab", totalLines: 5, totalBytes: 4000, truncated: true } },
+  };
+  const withoutOverflow = toolResultMessage("r2", "c2", "b".repeat(4_000));
+  const entries = [
+    entry("e1", assistantMessage("a1", [{ id: "c1", name: "bash" }, { id: "c2", name: "bash" }])),
+    entry("e2", withOverflow),
+    entry("e3", withoutOverflow),
+    ...protect(),
+  ];
+  const masking = { ...MASKING, minClearTokens: 1 };
+  const plan = planMasking({ contextEntries: entries, currentTokens: 3_500, contextWindow: 0, masking, estimate });
+  assert.equal(plan.masked.length, 2);
+  const first = plan.masked.find((m) => m.toolCallId === "c1")!.placeholder;
+  const second = plan.masked.find((m) => m.toolCallId === "c2")!.placeholder;
+  assert.ok(first.includes(` Full original stored: overflow_query {"id":"0123456789ab"}.`));
+  assert.ok(!second.includes("overflow_query"));
+});
+
+test("the small-window ratio caps both trigger and target", () => {
+  const masking: MaskingConfig = {
+    ...MASKING,
+    triggerTokens: 200_000,
+    triggerRatio: 0.6,
+    targetTokens: 100_000,
+    targetRatio: 0.35,
+    minClearTokens: 50_000,
+  };
+  assert.equal(effectiveThreshold(masking.triggerTokens, masking.triggerRatio, 272_000), 163_200);
+  assert.equal(effectiveThreshold(masking.targetTokens, masking.targetRatio, 272_000), 95_200);
+  const entries = [
+    entry("e1", toolResultMessage("r1", "c1", "a".repeat(300_000))),
+    ...protect(),
+  ];
+  const below = planMasking({ contextEntries: entries, currentTokens: 163_199, contextWindow: 272_000, masking, estimate });
+  assert.equal(below.drafts.length, 0);
+  const above = planMasking({ contextEntries: entries, currentTokens: 163_200, contextWindow: 272_000, masking, estimate });
+  assert.equal(above.drafts.length, 1);
+});
+
+test("ledger rebuild counts only masked targets still inside the active context", () => {
+  const placeholder = (id: string) =>
+    `${MASK_PREFIX}bash result cleared to save context (~100 tokens). Call args: {}. Re-run the call if still needed, or recover the original with context_tree_query {"toolCallIds":["${id}"]}.`;
+  const branch = [
+    { type: "message", id: "u1", parentId: null, timestamp: "t", message: { role: "user", content: "start", timestamp: 1 } },
+    { type: "message", id: "r-old", parentId: "u1", timestamp: "t", message: toolResultMessage("m-old", "c-old", "o".repeat(2_000)) },
+    { type: "context_edit", id: "e-old", parentId: "r-old", timestamp: "t", targetId: "r-old", replacement: { content: placeholder("c-old") } },
+    { type: "message", id: "m-keep", parentId: "e-old", timestamp: "t", message: { role: "user", content: "kept question", timestamp: 2 } },
+    { type: "compaction", id: "c1", parentId: "m-keep", timestamp: "t", summary: "summary", firstKeptEntryId: "m-keep", tokensBefore: 9_999 },
+    { type: "message", id: "a-keep", parentId: "c1", timestamp: "t", message: assistantMessage("a-keep", []) },
+    { type: "message", id: "r-new", parentId: "a-keep", timestamp: "t", message: toolResultMessage("m-new", "c-new", "n".repeat(2_000)) },
+    { type: "context_edit", id: "e-new", parentId: "r-new", timestamp: "t", targetId: "r-new", replacement: { content: placeholder("c-new") } },
+  ];
+  const ledger = rebuildMaskedLedger(branch, "e-new", estimate);
+  assert.equal(ledger.size, 1);
+  assert.ok(ledger.has("c-new"));
+  assert.ok(!ledger.has("c-old"));
+});
+
+test("user and assistant messages are never masked", () => {
+  const entries = [
+    entry("e1", { role: "user", content: "u".repeat(8_000), timestamp: "u1" }),
+    entry("e2", { role: "assistant", content: [{ type: "text", text: "a".repeat(8_000) }], timestamp: "a1" }),
+    ...protect(),
+  ];
+  const plan = planMasking({ contextEntries: entries, currentTokens: 9_000, contextWindow: 0, masking: { ...MASKING, keepRecentTurns: 1, minClearTokens: 1 }, estimate });
+  assert.equal(plan.drafts.length, 0);
+});

pi-compaction/test/performance.test.ts

@@ -10,9 +10,9 @@ function message(index: number) {
 
 test("cached projection and append-only delta stay inside the hot-path budget", () => {
   const index = new SegmentIndex("perf");
-  index.hydrate([], DEFAULT_CONFIG);
+  index.hydrate();
   const messages = Array.from({ length: 10_000 }, (_, value) => message(value));
-  const params = { environmentHash: "env", environmentTokens: 2_000, contextWindow: 300_000 };
+  const params = { environmentHash: "env", environmentTokens: 2_000, contextWindow: 300_000, maskedResults: 0, reclaimedTokens: 0 };
   index.project(messages, params);
 
   const cacheStart = performance.now();
@@ -24,6 +24,7 @@ test("cached projection and append-only delta stay inside the hot-path budget",
   index.project(appended, params);
   const deltaMs = performance.now() - deltaStart;
 
+  console.log(`cache-hit average ${cacheAverageMs.toFixed(3)}ms; delta projection ${deltaMs.toFixed(3)}ms`);
   assert.ok(cacheAverageMs < 2, `cache-hit average ${cacheAverageMs.toFixed(3)}ms exceeded 2ms`);
   assert.ok(deltaMs < 10, `delta projection ${deltaMs.toFixed(3)}ms exceeded 10ms`);
 });

pi-compaction/test/segment-index.test.ts

@@ -1,7 +1,6 @@
 import test from "node:test";
 import assert from "node:assert/strict";
-import { DEFAULT_CONFIG } from "../src/config.ts";
+import { SegmentIndex } from "../src/segment-index.ts";
-import { SegmentIndex, deterministicSummary } from "../src/segment-index.ts";
 
 const assistant = {
   role: "assistant",
@@ -17,66 +16,96 @@ const result = {
   timestamp: 11,
 };
 
-test("deterministic summary is bounded and preserves head/tail evidence", () => {
+// Regression: a ~14.5K-char read result projected under a 1M window must reach
-  const summary = deterministicSummary({
+// the model byte-for-byte unchanged — masking only happens via context_edit
-    toolName: "bash",
+// drafts under real pressure, never on the hot projection path.
-    args: { command: "npm test" },
+test("projection is byte-for-byte identity for a 300-line read result under a 1M window", () => {
-    rawText: result.content[0]!.text,
+  const readBody = Array.from({ length: 300 }, (_unused, index) => `line ${index + 1}: ${"y".repeat(38)}`).join("\n");
-    isError: true,
+  const readResult = {
-    maxStubTokens: 256,
+    role: "toolResult",
+    toolCallId: "call-read",
+    toolName: "read",
+    content: [{ type: "text", text: readBody }],
+    timestamp: 12,
+  };
+  assert.ok(readBody.length > 14_000 && readBody.length < 15_000, `fixture ${readBody.length} chars`);
+
+  const index = new SegmentIndex("session-1");
+  index.hydrate();
+  const snapshot = index.project([assistant, readResult], {
+    environmentHash: "env",
+    environmentTokens: 100,
+    contextWindow: 1_000_000,
+    maskedResults: 0,
+    reclaimedTokens: 0,
   });
-  assert.match(summary, /tool=bash status=error/);
+
-  assert.match(summary, /failure/);
+  assert.strictEqual(snapshot.messages[1], readResult);
-  assert.match(summary, /stack tail/);
+  assert.strictEqual(snapshot.messages[1]!.content[0].text, readBody);
-  assert.ok(summary.length <= 1_100);
+  assert.ok(snapshot.messages[1]!.content[0].text.includes("line 150:"));
+  assert.equal(snapshot.reclaimedTokens, 0);
+  assert.equal(snapshot.maskedResults, 0);
 });
 
-test("projection incrementally replaces tool results and caches identical input", () => {
+test("projection caches identical input and handles append-only deltas", () => {
-  const index = new SegmentIndex("session-1");
+  const index = new SegmentIndex("session-2");
-  index.hydrate([], DEFAULT_CONFIG);
+  index.hydrate();
-  const records = index.captureTurn(assistant, [result], DEFAULT_CONFIG);
-  assert.equal(records.length, 1);
-
   const messages = [assistant, result];
-  const first = index.project(messages, {
+  const params = { environmentHash: "env", environmentTokens: 100, contextWindow: 300_000, maskedResults: 0, reclaimedTokens: 0 };
-    environmentHash: "env",
+  const first = index.project(messages, params);
-    environmentTokens: 100,
+  const second = index.project(messages, params);
-    contextWindow: 300_000,
-  });
-  const second = index.project(messages, {
-    environmentHash: "env",
-    environmentTokens: 100,
-    contextWindow: 300_000,
-  });
-
   assert.strictEqual(first, second);
-  assert.ok(first.effectiveTokens < first.rawTokens);
+  assert.ok(first.rawTokens > 0);
-  assert.match(first.messages[1]!.content[0].text, /context_tree_query/);
 
-  const record = records[0]!;
+  const appended = index.project([...messages, { role: "user", content: "next", timestamp: 13 }], params);
-  assert.ok(index.applySemantic(record.sourceKey, record.sourceHash, "- npm test failed\n- stack tail"));
+  assert.notStrictEqual(appended, first);
-  const semantic = index.project(messages, {
+  assert.equal(appended.messages.length, 3);
+  assert.strictEqual(appended.messages[1], result);
+});
+
+test("changed masking stats publish a new snapshot without mutating the old one", () => {
+  const index = new SegmentIndex("session-stats");
+  index.hydrate();
+  const params = {
     environmentHash: "env",
     environmentTokens: 100,
     contextWindow: 300_000,
-  });
+    maskedResults: 0,
-  assert.notStrictEqual(semantic, first);
+    reclaimedTokens: 0,
-  assert.match(semantic.messages[1]!.content[0].text, /npm test failed/);
+  };
+  const first = index.project([assistant, result], params);
+  const same = index.project([assistant, result], params);
+  assert.strictEqual(same, first);
+
+  const updated = index.project([assistant, result], { ...params, maskedResults: 1, reclaimedTokens: 500 });
+  assert.notStrictEqual(updated, first);
+  assert.equal(updated.maskedResults, 1);
+  assert.equal(updated.reclaimedTokens, 500);
+  assert.equal(updated.rawTokens, updated.effectiveTokens + 500);
+  // The previously published object stays immutable.
+  assert.equal(first.maskedResults, 0);
+  assert.equal(first.reclaimedTokens, 0);
+  // Unchanged stats keep returning the same cached object.
+  assert.strictEqual(index.project([assistant, result], { ...params, maskedResults: 1, reclaimedTokens: 500 }), updated);
 });
 
 test("usage-backed baseline is adjusted only by projected delta", () => {
-  const index = new SegmentIndex("session-2");
+  const index = new SegmentIndex("session-3");
-  index.hydrate([], DEFAULT_CONFIG);
+  index.hydrate();
   const initial = index.project([assistant], {
     environmentHash: "env",
     environmentTokens: 50,
     contextWindow: 100_000,
+    maskedResults: 0,
+    reclaimedTokens: 0,
   });
   index.recordUsage(1_000, initial.messageEstimate, "env");
   const next = index.project([assistant, { role: "user", content: "next", timestamp: 12 }], {
     environmentHash: "env",
     environmentTokens: 50,
     contextWindow: 100_000,
+    maskedResults: 0,
+    reclaimedTokens: 0,
   });
   assert.equal(next.confidence, "usage-backed");
   assert.ok(next.effectiveTokens >= 1_000);

pi-compaction/test/token-accounting.test.ts

@@ -1,19 +1,17 @@
 import test from "node:test";
 import assert from "node:assert/strict";
-import { shouldCompact } from "../src/token-accounting.ts";
+import { effectiveThreshold } from "../src/token-accounting.ts";
 
-test("raw pressure cannot trigger compaction while effective lower-bound stays below 90%", () => {
+test("absolute token threshold wins on large windows", () => {
-  const config = { hardRatio: 0.9 };
+  assert.equal(effectiveThreshold(200_000, 0.6, 1_000_000), 200_000);
-  const snapshot = { lowerBound: 231_000, contextWindow: 300_000 };
-  assert.equal(shouldCompact("threshold", snapshot, config), false);
-  assert.equal(shouldCompact("manual", snapshot, config), true);
-  assert.equal(shouldCompact("overflow", snapshot, config), true);
 });
 
-test("effective lower-bound at the configured threshold compacts", () => {
+test("window ratio caps the threshold on small windows", () => {
-  assert.equal(shouldCompact(
+  assert.equal(effectiveThreshold(200_000, 0.6, 272_000), 163_200);
-    "threshold",
+  assert.equal(effectiveThreshold(100_000, 0.35, 272_000), 95_200);
-    { lowerBound: 270_000, contextWindow: 300_000 },
+});
-    { hardRatio: 0.9 },
+
-  ), true);
+test("a missing context window falls back to the absolute threshold", () => {
+  assert.equal(effectiveThreshold(200_000, 0.6, 0), 200_000);
+  assert.equal(effectiveThreshold(200_000, 0.6, -1), 200_000);
 });

pi-default-system-prompt.md

@@ -1,31 +0,0 @@
-You are an expert coding assistant operating inside pi, a coding agent harness. You help users by reading files, executing commands, editing code, and writing new files.
-
-Available tools:
-- read: Read file contents
-- bash: Execute bash commands (ls, grep, find, etc.)
-- edit: Make precise file edits with exact text replacement, including multiple disjoint edits in one call
-- write: Create or overwrite files
-
-In addition to the tools above, you may have access to other custom tools depending on the project.
-
-Guidelines:
-- Use bash for file operations like ls, rg, find
-- Use read to examine files instead of cat or sed.
-- You can inspect PI_* environment variables for current model and session details.
-- Use edit for precise changes (edits[].oldText must match exactly)
-- When changing multiple separate locations in one file, use one edit call with multiple entries in edits[] instead of multiple edit calls
-- Each edits[].oldText is matched against the original file, not after earlier edits are applied. Do not emit overlapping or nested edits. Merge nearby changes into one edit.
-- Keep edits[].oldText as small as possible while still being unique in the file. Do not pad with large unchanged regions.
-- Use write only for new files or complete rewrites.
-- Be concise in your responses
-- Show file paths clearly when working with files
-
-Pi documentation (read only when the user asks about pi itself, its SDK, extensions, themes, skills, or TUI):
-- Main documentation: /Users/sia/.nvm/versions/node/v24.16.0/lib/node_modules/@earendil-works/pi-coding-agent/README.md
-- Additional docs: /Users/sia/.nvm/versions/node/v24.16.0/lib/node_modules/@earendil-works/pi-coding-agent/docs
-- Examples: /Users/sia/.nvm/versions/node/v24.16.0/lib/node_modules/@earendil-works/pi-coding-agent/examples (extensions, custom tools, SDK)
-- When reading pi docs or examples, resolve docs/... under Additional docs and examples/... under Examples, not the current working directory
-- When asked about: extensions (docs/extensions.md, examples/extensions/), themes (docs/themes.md), skills (docs/skills.md), prompt templates (docs/prompt-templates.md), TUI components (docs/tui.md), keybindings (docs/keybindings.md), SDK integrations (docs/sdk.md), custom providers (docs/custom-provider.md), adding models (docs/models.md), pi packages (docs/packages.md), environment variables (docs/environment-variables.md)
-- When working on pi topics, read the docs and examples, and follow .md cross-references before implementing
-- Always read pi .md files completely and follow links to related docs (e.g., tui.md for TUI API details)
-Current working directory: /Users/sia/code/my-pi

pi-hermes-memory/UPSTREAM.md

@@ -19,7 +19,7 @@
 
 ### 依赖
 
-- `better-sqlite3@^13.0.3`：13.x 运行时实际 `dlopen` 的是包内 N-API 预编译二进制 `prebuilds/<platform>.node`，**不依赖 node-gyp 编译产物**。npm 会因该包存在 `binding.gyp` 而推断出隐式 `node-gyp rebuild`，但该步骤既无必要也不得批准。npm 会把它嵌套安装到本目录的 `node_modules/`，与 Context Mode 使用的根 `better-sqlite3@12.11.1` 并存；因此根 `allowScripts` 无需新增条目，也不得删除既有的 `better-sqlite3@12.11.1` 放行。
+- `better-sqlite3@^13.0.3`：13.x 运行时实际 `dlopen` 的是包内 N-API 预编译二进制 `prebuilds/<platform>.node`，**不依赖 node-gyp 编译产物**。npm 会因该包存在 `binding.gyp` 而推断出隐式 `node-gyp rebuild`，但该步骤既无必要也不得批准。npm 会把它嵌套安装到本目录的 `node_modules/`，根 `allowScripts` 无需任何 `better-sqlite3` 放行条目（context-mode 移除后根依赖已不再包含 better-sqlite3@12）。
 - `strip-ansi@7.2.0`：解析到根 `node_modules`。
 
 ### 本地改动

pi-minimal-footer/README.md

@@ -34,14 +34,14 @@ pi install npm:@ogulcancelik/pi-minimal-footer
 
 ## Configuration
 
-Environment variables (all optional):
+Persistent defaults are read from `config/pi-minimal-footer.json` inside the installed bundle; a missing or invalid file falls back to the defaults shown:
 
-| Variable                        | Description                                              | Default |
+| Field        | Description                                              | Default |
-| ------------------------------- | -------------------------------------------------------- | ------- |
+| ------------ | -------------------------------------------------------- | ------- |
-| `PI_MINIMAL_FOOTER_SHOW_CWD`    | Show current working directory in footer status line     | `1`     |
+| `showCwd`    | Show current working directory in footer status line     | `true`  |
-| `PI_MINIMAL_FOOTER_SHOW_BRANCH` | Show git branch/dirty/ahead/behind in footer status line | `1`     |
+| `showBranch` | Show git branch/dirty/ahead/behind in footer status line | `true`  |
 
-Accepted false values: `0`, `false`, `no`, `off` (case-insensitive).
+Only `true`/`false` values are honoured; anything else falls back to the default for that field.
 
 ## How it works
 

pi-minimal-footer/UPSTREAM.md

@@ -5,4 +5,4 @@
 - 导入包版本：`@ogulcancelik/pi-minimal-footer@0.1.10`
 - 导入日期：2026-08-19
 
-本目录由 `my-pi` 直接维护；它不是 submodule，也不保留上游 Git 元数据。本地改动新增了 Pi 扩展状态渲染（把 `ctx.ui.setStatus()` 发布的每个非空状态渲染为普通 footer 片段）。
+本目录由 `my-pi` 直接维护；它不是 submodule，也不保留上游 Git 元数据。本地改动新增了 Pi 扩展状态渲染（把 `ctx.ui.setStatus()` 发布的每个非空状态渲染为普通 footer 片段），并把 showCwd/showBranch 持久配置由 `PI_MINIMAL_FOOTER_*` 环境变量改为组合包 `config/pi-minimal-footer.json`。

pi-minimal-footer/index.ts

@@ -15,6 +15,7 @@ import { existsSync, readFileSync } from "node:fs";
 import { join } from "node:path";
 import { homedir } from "node:os";
 import { collectFooterStatusItems } from "./status.ts";
+import { readFooterSettings } from "./settings.ts";
 
 // ============ Types ============
 
@@ -43,19 +44,6 @@ interface GitCache {
 const USAGE_REFRESH_INTERVAL = 5 * 60_000; // 5 minutes
 const usageCache = new Map<string, UsageSnapshot>(); // keyed by provider
 
-// ============ Env Flags ============
-
-function parseBooleanEnv(value: string | undefined, fallback: boolean): boolean {
-  if (value === undefined) return fallback;
-
-  const normalized = value.trim().toLowerCase();
-  if (!normalized) return fallback;
-
-  if (["1", "true", "yes", "on"].includes(normalized)) return true;
-  if (["0", "false", "no", "off"].includes(normalized)) return false;
-  return fallback;
-}
-
 // ============ Identity Cache ============
 
 let codexEmailPrefix: string | null | undefined; // undefined = not yet resolved
@@ -754,9 +742,8 @@ export default function (pi: ExtensionAPI) {
   const BAR_FILLED = "━";
   const BAR_EMPTY = "─";
 
-  // Optional visibility toggles (default: enabled)
+  // Optional visibility toggles from bundle config (default: enabled)
-  const showCwd = parseBooleanEnv(process.env.PI_MINIMAL_FOOTER_SHOW_CWD, true);
+  const { showCwd, showBranch } = readFooterSettings();
-  const showBranch = parseBooleanEnv(process.env.PI_MINIMAL_FOOTER_SHOW_BRANCH, true);
 
   function formatTokenCount(tokens: number): string {
     if (tokens >= 1_000_000) {

pi-minimal-footer/settings.ts

@@ -0,0 +1,30 @@
+import { readFileSync } from "node:fs";
+import { fileURLToPath } from "node:url";
+
+export interface FooterSettings {
+  showCwd: boolean;
+  showBranch: boolean;
+}
+
+// Bundle-level `config/` is the single source of truth: the extension reads
+// `config/pi-minimal-footer.json` from inside the installed bundle (resolved
+// relative to this source file), never environment variables.
+export const FOOTER_CONFIG_PATH = fileURLToPath(
+  new URL("../config/pi-minimal-footer.json", import.meta.url),
+);
+
+export function readFooterSettings(configPath: string = FOOTER_CONFIG_PATH): FooterSettings {
+  try {
+    const parsed = JSON.parse(readFileSync(configPath, "utf8")) as unknown;
+    const raw =
+      parsed !== null && typeof parsed === "object" && !Array.isArray(parsed)
+        ? (parsed as Record<string, unknown>)
+        : {};
+    return {
+      showCwd: typeof raw.showCwd === "boolean" ? raw.showCwd : true,
+      showBranch: typeof raw.showBranch === "boolean" ? raw.showBranch : true,
+    };
+  } catch {
+    return { showCwd: true, showBranch: true };
+  }
+}

pi-minimal-footer/test/settings.test.ts

@@ -0,0 +1,51 @@
+import assert from "node:assert/strict";
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, describe, it } from "node:test";
+
+import { readFooterSettings } from "../settings.ts";
+
+let tempDir: string | undefined;
+
+afterEach(() => {
+  if (tempDir) {
+    rmSync(tempDir, { recursive: true, force: true });
+    tempDir = undefined;
+  }
+});
+
+function configPath(contents: string): string {
+  tempDir = mkdtempSync(join(tmpdir(), "pi-minimal-footer-settings-"));
+  const path = join(tempDir, "pi-minimal-footer.json");
+  writeFileSync(path, contents);
+  return path;
+}
+
+const expectedDefaults = { showCwd: true, showBranch: true };
+
+describe("readFooterSettings", () => {
+  it("returns all defaults when the config file is missing", () => {
+    assert.deepEqual(readFooterSettings("/does/not/exist/pi-minimal-footer.json"), expectedDefaults);
+  });
+
+  it("returns all defaults for malformed or non-object JSON", () => {
+    for (const contents of ["not json", "[1,2]", "42"]) {
+      assert.deepEqual(readFooterSettings(configPath(contents)), expectedDefaults, contents);
+    }
+  });
+
+  it("adopts explicit boolean values", () => {
+    assert.deepEqual(
+      readFooterSettings(configPath(JSON.stringify({ showCwd: false, showBranch: false }))),
+      { showCwd: false, showBranch: false },
+    );
+  });
+
+  it("falls back per-field for non-boolean values", () => {
+    assert.deepEqual(
+      readFooterSettings(configPath(JSON.stringify({ showCwd: "no", showBranch: 0 }))),
+      expectedDefaults,
+    );
+  });
+});

pi-notify/AGENTS.md

@@ -5,7 +5,7 @@
 - Kitty OSC 99 必须保留 source-window focus、Session 间唯一 ID、Base64 payload 和 tmux passthrough；修改 visibility/action、事件生命周期、fallback、模板、隐私或命令时必须同步 README 和测试。
 - 完成通知默认只在 TUI `agent_settled` 后发送；自动重试、自动 compaction 和 follow-up 期间不得提前通知或重置总耗时。权限通知只在 `permissions:ui_prompt` 后立即发送，不得为策略或 AutoReview 已直接决定的请求通知，也不受完成通知时长门槛影响。非 TUI 输出完整性、策略/AutoReview 静默决定、隐私模式、自动 retry/compaction/follow-up 去重、总耗时、短任务阈值和多窗口点击聚焦都属于回归边界。
 - 默认 `o=always`、`a=focus`，每个 Pi Session 使用独立稳定 ID，标题和正文必须 Base64 编码。
-- `/notify on|off` 只覆盖当前 Session；持久默认来自 `PI_NOTIFY_*` 环境变量。`PI_NOTIFY_MESSAGE_SOURCE=none` 必须继续提供不泄露回复正文、权限值或请求者细节的隐私模式，非 TUI 模式不得写入 OSC 序列。
+- `/notify on|off` 只覆盖当前 Session；持久默认来自组合包 `config/pi-notify.json`（`src/settings.ts` 读取，文件缺失或字段非法逐字段回退默认）。`messageSource: "none"` 必须继续提供不泄露回复正文、权限值或请求者细节的隐私模式，非 TUI 模式不得写入 OSC 序列。
 - 根包必须通过 `file:./pi-notify` 提供依赖并直接加载源码，不得同时加载 npm `@smoose/pi-notify` 实现，否则会重复通知。
 
 ## 验证

pi-notify/README.md

@@ -28,19 +28,21 @@ Do not separately install upstream when using the bundle, or notifications will
 - `/notify test`: send a notification immediately; click it to verify exact Kitty-window focus.
 - `/notify status`: show the active backend, visibility policy, and duration threshold.
 
-Command overrides are session-local. Use environment variables for persistent defaults.
+Command overrides are session-local. Persistent defaults come from the bundle's `config/pi-notify.json`.
 
 ## Configuration
 
-- `PI_NOTIFY_ENABLED`: `1`/`true` or `0`/`false`; default enabled.
+Persistent defaults are read once per session from `config/pi-notify.json` inside the installed bundle; a missing or invalid file falls back to the defaults shown:
-- `PI_NOTIFY_TITLE`: title template shared by completion and permission notifications; default `π - {project}`.
+
-- `PI_NOTIFY_BODY`: completion-notification body template; default `{message}`. Permission prompts use a bounded `Permission required · <agent> · <surface>` summary.
+- `enabled`: `true`/`false`; default `true`.
-- `PI_NOTIFY_MESSAGE_SOURCE`: `assistant`, `user`, or privacy mode `none`; default `assistant`. Privacy mode also hides permission values and requester details.
+- `title`: title template shared by completion and permission notifications; default `π - {project}`.
-- `PI_NOTIFY_MESSAGE_MAX`: maximum completion-message or permission-value characters before ellipsis; default `80`, maximum `500`.
+- `body`: completion-notification body template; default `{message}`. Permission prompts use a bounded `Permission required · <agent> · <surface>` summary.
-- `PI_NOTIFY_MIN_SECONDS`: minimum complete run duration; default `3`, range `0`–`3600`. This does not delay permission prompts.
+- `messageSource`: `assistant`, `user`, or privacy mode `none`; default `assistant`. Privacy mode also hides permission values and requester details.
-- `PI_NOTIFY_VISIBILITY`: Kitty policy `unfocused`, `invisible`, or `always`; default `always`.
+- `messageMax`: maximum completion-message or permission-value characters before ellipsis; default `80`, maximum `500`.
-- `PI_NOTIFY_ACTION`: Kitty click action `focus` or `none`; default `focus`.
+- `minSeconds`: minimum complete run duration; default `3`, range `0`–`3600`. This does not delay permission prompts.
-- `PI_NOTIFY_SOUND_CMD`: optional shell command run after the notification. This is trusted local configuration and executes through a shell.
+- `visibility`: Kitty policy `unfocused`, `invisible`, or `always`; default `always`.
+- `action`: Kitty click action `focus` or `none`; default `focus`.
+- `soundCommand`: optional shell command run after the notification. This is trusted local configuration and executes through a shell.
 
 Visibility policies:
 
@@ -70,12 +72,14 @@ The active tmux client type is checked so a stale `KITTY_WINDOW_ID` cannot force
 - `{status_icon}`: `✓` or `⚠`. This reports observed tool errors, not a definitive task-success verdict.
 - `{tool_error_count}`: observed tool error count.
 
-Example for multiple projects without exposing assistant text on the lock screen:
+Example for multiple projects without exposing assistant text on the lock screen (`config/pi-notify.json`):
 
-```bash
+```json
-export PI_NOTIFY_TITLE='π {status_icon} · {project}'
+{
-export PI_NOTIFY_MESSAGE_SOURCE=none
+  "title": "π {status_icon} · {project}",
-export PI_NOTIFY_VISIBILITY=unfocused
+  "messageSource": "none",
+  "visibility": "unfocused"
+}
 ```
 
 Kitty must be allowed under **System Settings → Notifications**. Self-built or unsigned Kitty binaries may not be able to publish macOS notifications.

pi-notify/UPSTREAM.md

@@ -18,5 +18,6 @@
 - 新增时长过滤、启用/隐私/动作设置、状态模板字段和 `/notify` 控制。
 - 保留 Base64 Kitty payload，加固 OSC 元数据/回退净化，并忽略分离进程的 spawn 错误。
 - 新增生命周期、多窗口元数据、回退、解析器和配置测试。
+- 持久配置由 `PI_NOTIFY_*` 环境变量改为组合包 `config/pi-notify.json`（`src/settings.ts` 读取并逐字段回退默认值）。
 
 上游声明本项目使用 MIT 许可证，但在导入的 commit 和 npm tarball 中都没有包含 `LICENSE` 文件。本目录使用上游作者的 GitHub 身份和 commit 年份补充标准 MIT 文本，使声明的许可证随维护源码一起保留。

pi-notify/src/config.ts

@@ -1,13 +1,6 @@
 import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
 import { extractMessageText, truncateText } from "./messages.ts";
-import {
+import { readNotifySettings, type NotifySettings } from "./settings.ts";
-  parseEnabled,
-  parseKittyVisibility,
-  parseMessageMax,
-  parseMessageSource,
-  parseMinDurationMs,
-  parseNotificationAction,
-} from "./parsers.ts";
 import { renderTemplate } from "./template.ts";
 import type { NotificationRuntimeValues, NotifyConfig } from "./types.ts";
 
@@ -20,12 +13,13 @@ export function getConfig(
   messages: unknown[] | undefined,
   runtime: NotificationRuntimeValues,
   now = Date.now(),
+  settings: NotifySettings = readNotifySettings(),
 ): NotifyConfig {
   const durationMs = startedAt === undefined ? 0 : Math.max(0, now - startedAt);
-  const messageMax = parseMessageMax(process.env.PI_NOTIFY_MESSAGE_MAX);
+  const messageMax = settings.messageMax;
   const userMessage = truncateText(extractMessageText(messages, "user"), messageMax);
   const assistantMessage = truncateText(extractMessageText(messages, "assistant"), messageMax);
-  const messageSource = parseMessageSource(process.env.PI_NOTIFY_MESSAGE_SOURCE);
+  const messageSource = settings.messageSource;
   const selectedMessage = messageSource === "user" ? userMessage : messageSource === "assistant" ? assistantMessage : "";
   const message = selectedMessage || "Ready for input";
   const toolErrorCount = Math.max(0, runtime.toolErrorCount ?? 0);
@@ -39,15 +33,15 @@ export function getConfig(
   };
 
   return {
-    enabled: parseEnabled(process.env.PI_NOTIFY_ENABLED),
+    enabled: settings.enabled,
-    title: renderTemplate(process.env.PI_NOTIFY_TITLE || DEFAULT_TITLE, ctx, durationMs, values),
+    title: renderTemplate(settings.title || DEFAULT_TITLE, ctx, durationMs, values),
-    body: renderTemplate(process.env.PI_NOTIFY_BODY || DEFAULT_BODY, ctx, durationMs, values),
+    body: renderTemplate(settings.body || DEFAULT_BODY, ctx, durationMs, values),
     durationMs,
-    minDurationMs: parseMinDurationMs(process.env.PI_NOTIFY_MIN_SECONDS),
+    minDurationMs: settings.minDurationMs,
     notificationId: runtime.notificationId,
-    visibility: parseKittyVisibility(process.env.PI_NOTIFY_VISIBILITY),
+    visibility: settings.visibility,
-    action: parseNotificationAction(process.env.PI_NOTIFY_ACTION),
+    action: settings.action,
-    soundCommand: process.env.PI_NOTIFY_SOUND_CMD?.trim() || undefined,
+    soundCommand: settings.soundCommand,
   };
 }
 

pi-notify/src/index.ts

@@ -2,7 +2,7 @@ import { randomUUID } from "node:crypto";
 import type { ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
 import { getConfig, shouldNotify } from "./config.ts";
 import { truncateText } from "./messages.ts";
-import { parseMessageMax, parseMessageSource } from "./parsers.ts";
+import { readNotifySettings, type NotifySettings } from "./settings.ts";
 import { getNotificationBackend, sendNotification } from "./terminal.ts";
 import type { NotifyConfig } from "./types.ts";
 
@@ -11,13 +11,14 @@ type ExtensionDependencies = {
   createNotificationId: () => string;
   notify: (config: NotifyConfig) => void;
   getBackend: typeof getNotificationBackend;
+  settings: NotifySettings;
 };
 
 const PERMISSIONS_UI_PROMPT_CHANNEL = "permissions:ui_prompt";
 
-function formatPermissionPromptBody(raw: unknown): string | undefined {
+function formatPermissionPromptBody(raw: unknown, settings: NotifySettings): string | undefined {
   if (!isRecord(raw) || !readNonEmptyString(raw.requestId)) return undefined;
-  if (parseMessageSource(process.env.PI_NOTIFY_MESSAGE_SOURCE) === "none") {
+  if (settings.messageSource === "none") {
     return "Permission confirmation required";
   }
 
@@ -29,7 +30,7 @@ function formatPermissionPromptBody(raw: unknown): string | undefined {
 
   if (!value) return labels.join(" · ");
   const normalizedValue = value.replace(/\s+/g, " ").trim();
-  return `${labels.join(" · ")}\n${truncateText(normalizedValue, parseMessageMax(process.env.PI_NOTIFY_MESSAGE_MAX))}`;
+  return `${labels.join(" · ")}\n${truncateText(normalizedValue, settings.messageMax)}`;
 }
 
 function readNonEmptyString(value: unknown): string | undefined {
@@ -47,6 +48,7 @@ export function createPiNotifyExtension(dependencies: Partial<ExtensionDependenc
   const createNotificationId = dependencies.createNotificationId ?? (() => `pi-${process.pid}-${randomUUID()}`);
   const notify = dependencies.notify ?? sendNotification;
   const getBackend = dependencies.getBackend ?? getNotificationBackend;
+  const settings = dependencies.settings ?? readNotifySettings();
 
   return function piNotify(pi: ExtensionAPI): void {
     const notificationId = createNotificationId();
@@ -81,7 +83,7 @@ export function createPiNotifyExtension(dependencies: Partial<ExtensionDependenc
 
     pi.on("agent_settled", (_event, ctx) => {
       try {
-        const config = getConfig(ctx, startedAt, latestMessages, { notificationId, toolErrorCount }, now());
+        const config = getConfig(ctx, startedAt, latestMessages, { notificationId, toolErrorCount }, now(), settings);
         if (ctx.mode === "tui" && shouldNotify(config, enabledOverride)) notify(config);
       } finally {
         resetRun();
@@ -89,7 +91,7 @@ export function createPiNotifyExtension(dependencies: Partial<ExtensionDependenc
     });
 
     const unsubscribePermissionPrompt = pi.events.on(PERMISSIONS_UI_PROMPT_CHANNEL, (raw) => {
-      const body = formatPermissionPromptBody(raw);
+      const body = formatPermissionPromptBody(raw, settings);
       if (!body || currentCtx?.mode !== "tui") return;
 
       const config = getConfig(
@@ -98,6 +100,7 @@ export function createPiNotifyExtension(dependencies: Partial<ExtensionDependenc
         undefined,
         { notificationId, toolErrorCount },
         now(),
+        settings,
       );
       const permissionConfig = { ...config, body, minDurationMs: 0 };
       if (shouldNotify(permissionConfig, enabledOverride)) notify(permissionConfig);
@@ -119,7 +122,7 @@ export function createPiNotifyExtension(dependencies: Partial<ExtensionDependenc
           return;
         }
 
-        const config = getConfig(ctx, now() - 1000, undefined, { notificationId, toolErrorCount: 0 }, now());
+        const config = getConfig(ctx, now() - 1000, undefined, { notificationId, toolErrorCount: 0 }, now(), settings);
         if (action === "test") {
           if (ctx.mode !== "tui") {
             ctx.ui.notify("System notifications are only emitted in interactive TUI mode", "warning");

pi-notify/src/settings.ts

@@ -0,0 +1,63 @@
+import { readFileSync } from "node:fs";
+import { fileURLToPath } from "node:url";
+import {
+  parseEnabled,
+  parseKittyVisibility,
+  parseMessageMax,
+  parseMessageSource,
+  parseMinDurationMs,
+  parseNotificationAction,
+} from "./parsers.ts";
+import type { KittyVisibility, MessageSource, NotificationAction } from "./types.ts";
+
+export interface NotifySettings {
+  enabled: boolean;
+  title: string;
+  body: string;
+  messageSource: MessageSource;
+  messageMax: number;
+  minDurationMs: number;
+  visibility: KittyVisibility;
+  action: NotificationAction;
+  soundCommand?: string;
+}
+
+const DEFAULT_TITLE = "π - {project}";
+const DEFAULT_BODY = "{message}";
+
+// Bundle-level `config/` is the single source of truth: the extension reads
+// `config/pi-notify.json` from inside the installed bundle (resolved relative
+// to this source file), never environment variables.
+export const NOTIFY_CONFIG_PATH = fileURLToPath(new URL("../../config/pi-notify.json", import.meta.url));
+
+function readConfigFile(configPath: string): Record<string, unknown> {
+  try {
+    const parsed = JSON.parse(readFileSync(configPath, "utf8")) as unknown;
+    return parsed !== null && typeof parsed === "object" && !Array.isArray(parsed)
+      ? (parsed as Record<string, unknown>)
+      : {};
+  } catch {
+    return {};
+  }
+}
+
+function optionalString(value: unknown): string | undefined {
+  if (typeof value === "string") return value;
+  if (typeof value === "number" || typeof value === "boolean") return String(value);
+  return undefined;
+}
+
+export function readNotifySettings(configPath: string = NOTIFY_CONFIG_PATH): NotifySettings {
+  const raw = readConfigFile(configPath);
+  return {
+    enabled: parseEnabled(optionalString(raw.enabled)),
+    title: typeof raw.title === "string" && raw.title ? raw.title : DEFAULT_TITLE,
+    body: typeof raw.body === "string" && raw.body ? raw.body : DEFAULT_BODY,
+    messageSource: parseMessageSource(optionalString(raw.messageSource)),
+    messageMax: parseMessageMax(optionalString(raw.messageMax)),
+    minDurationMs: parseMinDurationMs(optionalString(raw.minSeconds)),
+    visibility: parseKittyVisibility(optionalString(raw.visibility)),
+    action: parseNotificationAction(optionalString(raw.action)),
+    soundCommand: typeof raw.soundCommand === "string" ? raw.soundCommand.trim() || undefined : undefined,
+  };
+}

pi-notify/tests/config.test.ts

@@ -1,28 +1,18 @@
 import assert from "node:assert/strict";
-import { afterEach, describe, it } from "node:test";
+import { describe, it } from "node:test";
 import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
 import { getConfig, shouldNotify } from "../src/config.ts";
+import type { NotifySettings } from "../src/settings.ts";
 
-const envKeys = [
+const defaultSettings = (): NotifySettings => ({
-  "PI_NOTIFY_ENABLED",
+  enabled: true,
-  "PI_NOTIFY_TITLE",
+  title: "π - {project}",
-  "PI_NOTIFY_BODY",
+  body: "{message}",
-  "PI_NOTIFY_MESSAGE_SOURCE",
+  messageSource: "assistant",
-  "PI_NOTIFY_MESSAGE_MAX",
+  messageMax: 80,
-  "PI_NOTIFY_MIN_SECONDS",
+  minDurationMs: 3_000,
-  "PI_NOTIFY_VISIBILITY",
+  visibility: "always",
-  "PI_NOTIFY_ACTION",
+  action: "focus",
-  "PI_NOTIFY_SOUND_CMD",
-] as const;
-
-const savedEnv = new Map(envKeys.map((key) => [key, process.env[key]]));
-
-afterEach(() => {
-  for (const key of envKeys) {
-    const value = savedEnv.get(key);
-    if (value === undefined) delete process.env[key];
-    else process.env[key] = value;
-  }
 });
 
 function context(): ExtensionContext {
@@ -45,6 +35,7 @@ describe("config", () => {
       ],
       runtime,
       5_000,
+      defaultSettings(),
     );
 
     assert.deepEqual(config, {
@@ -62,19 +53,20 @@ describe("config", () => {
   });
 
   it("supports privacy, threshold, focus, and status template settings", () => {
-    process.env.PI_NOTIFY_ENABLED = "off";
-    process.env.PI_NOTIFY_TITLE = "{status_icon} {status} {tool_error_count}";
-    process.env.PI_NOTIFY_BODY = "{message}";
-    process.env.PI_NOTIFY_MESSAGE_SOURCE = "none";
-    process.env.PI_NOTIFY_MIN_SECONDS = "10.5";
-    process.env.PI_NOTIFY_VISIBILITY = "invisible";
-    process.env.PI_NOTIFY_ACTION = "none";
-    process.env.PI_NOTIFY_SOUND_CMD = " say done ";
-
     const config = getConfig(context(), 1_000, [{ role: "assistant", content: "secret" }], {
       notificationId: "pi-session",
       toolErrorCount: 2,
-    }, 2_000);
+    }, 2_000, {
+      ...defaultSettings(),
+      enabled: false,
+      title: "{status_icon} {status} {tool_error_count}",
+      body: "{message}",
+      messageSource: "none",
+      minDurationMs: 10_500,
+      visibility: "invisible",
+      action: "none",
+      soundCommand: "say done",
+    });
 
     assert.equal(config.title, "⚠ ready · 2 tool errors 2");
     assert.equal(config.body, "Ready for input");
@@ -88,15 +80,13 @@ describe("config", () => {
   });
 
   it("applies selected-message truncation and a session override", () => {
-    process.env.PI_NOTIFY_MESSAGE_SOURCE = "user";
-    process.env.PI_NOTIFY_MESSAGE_MAX = "5";
-    process.env.PI_NOTIFY_MIN_SECONDS = "0";
     const config = getConfig(
       context(),
       1_000,
       [{ role: "user", content: "hello world" }],
       runtime,
       2_000,
+      { ...defaultSettings(), messageSource: "user", messageMax: 5, minDurationMs: 0 },
     );
 
     assert.equal(config.body, "hell…");

pi-notify/tests/index.test.ts

@@ -1,23 +1,24 @@
 import assert from "node:assert/strict";
-import { afterEach, describe, it } from "node:test";
+import { describe, it } from "node:test";
 import type { ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
 import { createPiNotifyExtension } from "../src/index.ts";
+import type { NotifySettings } from "../src/settings.ts";
 import type { NotifyConfig } from "../src/types.ts";
 
 type Handler = (...args: any[]) => any;
 
-const savedEnv = new Map(
+const DEFAULT_SETTINGS: NotifySettings = {
-  ["PI_NOTIFY_MIN_SECONDS", "PI_NOTIFY_MESSAGE_SOURCE"].map((key) => [key, process.env[key]]),
+  enabled: true,
-);
+  title: "π - {project}",
+  body: "{message}",
+  messageSource: "assistant",
+  messageMax: 80,
+  minDurationMs: 3_000,
+  visibility: "always",
+  action: "focus",
+};
 
-afterEach(() => {
+function harness(mode: ExtensionContext["mode"] = "tui", settings: Partial<NotifySettings> = {}) {
-  for (const [key, value] of savedEnv) {
-    if (value === undefined) delete process.env[key];
-    else process.env[key] = value;
-  }
-});
-
-function harness(mode: ExtensionContext["mode"] = "tui") {
   const handlers = new Map<string, Handler>();
   const eventHandlers = new Map<string, Handler>();
   const commands = new Map<string, { handler: Handler }>();
@@ -45,6 +46,7 @@ function harness(mode: ExtensionContext["mode"] = "tui") {
     createNotificationId: () => "pi-session-test",
     notify: (config) => notifications.push(config),
     getBackend: () => "kitty",
+    settings: { ...DEFAULT_SETTINGS, ...settings },
   })(pi);
 
   const ctx = {
@@ -93,9 +95,8 @@ describe("pi-notify extension", () => {
   });
 
   it("does not write terminal notifications in RPC or non-interactive modes", async () => {
-    process.env.PI_NOTIFY_MIN_SECONDS = "0";
     for (const mode of ["rpc", "json", "print"] as const) {
-      const h = harness(mode);
+      const h = harness(mode, { minDurationMs: 0 });
       await h.handlers.get("session_start")?.({}, h.ctx);
       await h.handlers.get("agent_start")?.({}, h.ctx);
       await h.handlers.get("agent_settled")?.({}, h.ctx);
@@ -123,8 +124,7 @@ describe("pi-notify extension", () => {
   });
 
   it("notifies immediately when permission confirmation needs human input", async () => {
-    process.env.PI_NOTIFY_MIN_SECONDS = "3600";
+    const h = harness("tui", { minDurationMs: 3_600_000 });
-    const h = harness();
     await h.handlers.get("session_start")?.({}, h.ctx);
 
     h.eventHandlers.get("permissions:ui_prompt")?.({
@@ -144,8 +144,7 @@ describe("pi-notify extension", () => {
   });
 
   it("keeps permission details private and honors the session off override", async () => {
-    process.env.PI_NOTIFY_MESSAGE_SOURCE = "none";
+    const h = harness("tui", { messageSource: "none" });
-    const h = harness();
     await h.handlers.get("session_start")?.({}, h.ctx);
     const prompt = { requestId: "perm-secret", surface: "bash", value: "cat private-file" };
 

pi-notify/tests/settings.test.ts

@@ -0,0 +1,107 @@
+import assert from "node:assert/strict";
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, describe, it } from "node:test";
+import { readNotifySettings } from "../src/settings.ts";
+
+let tempDir: string | undefined;
+
+afterEach(() => {
+  if (tempDir) {
+    rmSync(tempDir, { recursive: true, force: true });
+    tempDir = undefined;
+  }
+});
+
+function configPath(contents: string): string {
+  tempDir = mkdtempSync(join(tmpdir(), "pi-notify-settings-"));
+  const path = join(tempDir, "pi-notify.json");
+  writeFileSync(path, contents);
+  return path;
+}
+
+const expectedDefaults = {
+  enabled: true,
+  title: "π - {project}",
+  body: "{message}",
+  messageSource: "assistant",
+  messageMax: 80,
+  minDurationMs: 3_000,
+  visibility: "always",
+  action: "focus",
+  soundCommand: undefined,
+};
+
+describe("readNotifySettings", () => {
+  it("returns all defaults when the config file is missing", () => {
+    assert.deepEqual(readNotifySettings("/does/not/exist/pi-notify.json"), expectedDefaults);
+  });
+
+  it("returns all defaults for malformed or non-object JSON", () => {
+    for (const contents of ["not json", "[1,2]", "\"hi\"", "42"]) {
+      assert.deepEqual(readNotifySettings(configPath(contents)), expectedDefaults, contents);
+    }
+  });
+
+  it("adopts valid values from the config file", () => {
+    const settings = readNotifySettings(configPath(JSON.stringify({
+      enabled: false,
+      title: "π {status_icon} · {project}",
+      body: "{status}",
+      messageSource: "none",
+      messageMax: 120,
+      minSeconds: 30,
+      visibility: "unfocused",
+      action: "none",
+      soundCommand: "afplay /done.aiff",
+    })));
+
+    assert.deepEqual(settings, {
+      enabled: false,
+      title: "π {status_icon} · {project}",
+      body: "{status}",
+      messageSource: "none",
+      messageMax: 120,
+      minDurationMs: 30_000,
+      visibility: "unfocused",
+      action: "none",
+      soundCommand: "afplay /done.aiff",
+    });
+  });
+
+  it("falls back per-field when values have the wrong type or value", () => {
+    const settings = readNotifySettings(configPath(JSON.stringify({
+      enabled: {},
+      title: "",
+      body: null,
+      messageSource: ["assistant"],
+      messageMax: "bogus",
+      minSeconds: "bogus",
+      visibility: "bogus",
+      action: {},
+    })));
+
+    assert.deepEqual(settings, expectedDefaults);
+  });
+
+  it("clamps messageMax and minSeconds to their bounded ranges", () => {
+    const settings = readNotifySettings(configPath(JSON.stringify({ messageMax: 999, minSeconds: 4000 })));
+    assert.equal(settings.messageMax, 500);
+    assert.equal(settings.minDurationMs, 3_600_000);
+
+    const floor = readNotifySettings(configPath(JSON.stringify({ messageMax: -5, minSeconds: -1 })));
+    assert.equal(floor.messageMax, 0);
+    assert.equal(floor.minDurationMs, 0);
+  });
+
+  it("treats an empty or whitespace soundCommand as absent and trims values", () => {
+    for (const value of ["", "   "]) {
+      assert.equal(readNotifySettings(configPath(JSON.stringify({ soundCommand: value }))).soundCommand, undefined);
+    }
+    assert.equal(
+      readNotifySettings(configPath(JSON.stringify({ soundCommand: "  afplay /done.aiff  " }))).soundCommand,
+      "afplay /done.aiff",
+    );
+  });
+});

pi-overflow/AGENTS.md

@@ -0,0 +1,18 @@
+# my-pi 维护约定（pi-overflow）
+
+- 完全自维护（无上游）的新结果展示截断扩展，负责"每个新 tool_result 给模型看多少"：`capture.ts` 最先加载只暂存原文，`index.ts` 最后加载（在 `system-prompt.ts` 之前）生成有界视图并注册 `overflow_query`。
+- 根 `pi.extensions` 顺序是硬约束：`./pi-overflow/capture.ts` 必须第一，`./pi-overflow/index.ts` 必须在所有其他扩展之后、`./extensions/system-prompt.ts` 之前；跨扩展暂存走 `globalThis[Symbol.for("my-pi.pi-overflow.stash.v1")]`，容量 64 条、淘汰最旧。
+- `tool_result` handler 链中 capture 不得修改结果；index 改写 content 时必须原样回传 `structuredContent`、保留 image 部分、合并而非丢弃已有 `details`（如 RTK 压缩元数据、bash `fullOutputPath`）。
+- `read`/`edit`/`write`/`overflow_query`/`ask_user_question` 与嵌套调用（`parentToolCallId`）永远豁免；read 靠自身 offset/limit 分页，不得破坏 hashline 锚点。
+- 视图与行号一律基于刚落盘的存储原文（不是 current 文本）：文件来源用流式逐行扫描收集 head/环形 tail/signals，文本来源切同构片段；视图必须确定性、不超 `thresholdChars`、结尾永远可见；落盘原文优先取 `details.fullOutputPath` 文件（超 `maxEntryBytes` 只留头尾各半），其次暂存原文，再次当前文本。
+- 超预算收缩优先级为 tail > signals > head：head 先缩到 min(10, headLines)，signals 逐步减半至 10 再到 0，再按约 35/65 缩 head/tail，按字符截断兜底。
+- signal 扫描与 `overflow_query` 的 search 正则只匹配每行前 10,000 字符；query 输出每行按 `maxLineChars` 截断。
+- 存储位于 `<agentDir>/my-pi/pi-overflow/`（目录 `700`、文件 `600`、原子写入），id 为 `sha1(sessionId + "\0" + toolCallId)` 前 12 位 hex；`overflow_query` 的 id 必须匹配 `^[0-9a-f]{12}$` 防路径穿越；`session_start` 按 `retentionDays`/`maxStoreBytes` 清理并删除超过 1 小时的 `*.tmp-*` 残留，失败不得抛出。
+- 持久配置只读组合包 `config/pi-overflow.json`（`settings.ts` 逐字段校验、非法回退默认、signalPattern 编译失败回退默认且不敏感大小写）；不得读写 `settings.json` 或以环境变量作持久默认。
+- `overflow_query` 只读本扩展自己的存储目录；非法 id/正则/JSON/路径一律返回清楚错误文本，不得抛出。
+- 与 pi-compaction 的唯一联动是被遮蔽 toolResult 的 `details.overflow.id` 公开字段——pi-compaction 只读它、不得 import 本包任何实现。
+- 测试必须注入临时存储目录，不得写真实 agent 目录。
+
+## 验证
+
+- 在本目录内至少运行 `npm test` 与 `npm run check`；涉及加载入口、`pi.extensions` 顺序或 `files` 白名单时还要运行根 `node --test tests/extension-load.test.ts` 与 `tests/package-files.test.ts`。

pi-overflow/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 sia
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

pi-overflow/README.md

@@ -0,0 +1,66 @@
+# pi-overflow
+
+Bounded display for oversized tool results. When a tool result exceeds
+`thresholdChars`, pi-overflow persists the full original to disk and shows the
+model a deterministic view — head lines, signal lines (error/fail/warn/…), and
+tail lines — plus an `overflow_query` tool to search or read the stored
+original by id.
+
+## Two-stage load order
+
+- `capture.ts` — **first** in `pi.extensions`. A `tool_result` handler that
+  never modifies anything; it stashes the untouched original text in
+  `globalThis[Symbol.for("my-pi.pi-overflow.stash.v1")]` (capacity 64) so the
+  bound stage can still reach the pre-RTK original.
+- `index.ts` — **last** content extension, right before `system-prompt.ts`.
+  Its `tool_result` handler sees the final text after every other handler
+  (RTK semantic compaction included) has run, then:
+  - resolves the *original*: `details.fullOutputPath` file (streamed, capped
+    at `maxEntryBytes` head+tail) → stashed raw text → current text;
+  - if the current text still exceeds `thresholdChars`, replaces it with the
+    bounded view **built from the stored file** — line numbers and totals in
+    the view are exactly what `overflow_query` will read, even when RTK or
+    bash already cut the in-context text (file sources are scanned
+    line-by-line in a stream, never held in memory);
+  - if the current text fits but the original did not (an earlier extension
+    already cut it), keeps the text and appends a one-line
+    `[pi-overflow id=…]` recovery marker;
+  - preserves image parts and `structuredContent`, and merges
+    `details.overflow = { id, totalLines, totalBytes, truncated }` into the
+    existing details.
+- Exempt tools (`read`, `edit`, `write`, `overflow_query`,
+  `ask_user_question`) and nested calls (`parentToolCallId`) are never
+  touched.
+
+## Stored originals
+
+`<agentDir>/my-pi/pi-overflow/<id>.txt` + `<id>.json` metadata, where
+`id = sha1(sessionId + "\0" + toolCallId)[0:12]`. Directories are `0700`,
+files `0600`, writes are atomic. `session_start` cleans entries older than
+`retentionDays` and evicts oldest-first past `maxStoreBytes`.
+
+## `overflow_query`
+
+- `id` (required, `^[0-9a-f]{12}$` — path traversal safe)
+- `search` — case-insensitive regex; hits come back with `context` lines and
+  adjacent hits merge; capped by `maxResults`
+- `lines: { start, end }` — 1-based inclusive line range
+- `jsonPath` — dot/bracket path into stored JSON (`a.b[3].c`)
+- No selector → metadata + first-screen view. Output is always bounded by
+  `thresholdChars` and reports whether it was truncated.
+
+## Configuration
+
+`config/pi-overflow.json` inside the installed bundle. Every field is
+validated independently and falls back to its default; an uncompilable
+`signalPattern` falls back to the default pattern (case-insensitive).
+
+## Development
+
+```sh
+npm test        # node --test test/*.test.ts
+npm run check   # strip-types syntax check + tests
+```
+
+Tests inject a temporary store root — they never write to a real agent
+directory.

pi-overflow/UPSTREAM.md

@@ -0,0 +1,8 @@
+# 上游来源
+
+- 本项目完全自维护，无上游；未从任何外部仓库或 npm 包导入源码。
+- 实现仅依赖 Pi 扩展宿主 API（`@earendil-works/pi-coding-agent`）与 `typebox`，两者均为组合包 peer 依赖。
+
+## 本地改动记录
+
+- 初始实现：`capture.ts`（首个加载，暂存原始 tool_result 文本）、`index.ts`（最后加载，生成有界视图 + 注册 `overflow_query`）、`src/`（视图、存储、查询、跨扩展暂存区）。

pi-overflow/capture.ts

@@ -0,0 +1,30 @@
+import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
+import { readOverflowSettings } from "./settings.ts";
+import { getStash, stashResult } from "./src/stash.ts";
+
+/**
+ * Pre-bound capture: loaded FIRST in pi.extensions, before RTK or any other
+ * result-rewriting extension. It only stashes the untouched original text —
+ * it never modifies the result. The bound stage (`index.ts`, loaded last)
+ * consumes the stash after all other handlers have run.
+ */
+export default function overflowCapture(pi: ExtensionAPI): void {
+	const settings = readOverflowSettings();
+	pi.on("tool_result", (event) => {
+		if (!settings.enabled) return undefined;
+		if (settings.exemptTools.includes(event.toolName)) return undefined;
+		if (event.parentToolCallId) return undefined;
+		const rawText = event.content
+			.filter((part): part is { type: "text"; text: string } => part.type === "text")
+			.map((part) => part.text)
+			.join("\n");
+		const fullOutputPath =
+			typeof event.details?.fullOutputPath === "string" ? event.details.fullOutputPath : undefined;
+		if (rawText.length <= settings.thresholdChars && !fullOutputPath) return undefined;
+		stashResult(getStash(), event.toolCallId, {
+			rawText,
+			...(fullOutputPath ? { fullOutputPath } : {}),
+		});
+		return undefined;
+	});
+}

pi-overflow/index.ts

@@ -0,0 +1,167 @@
+import { getAgentDir, type ExtensionAPI } from "@earendil-works/pi-coding-agent";
+import { Type } from "typebox";
+import { stat } from "node:fs/promises";
+import { join } from "node:path";
+import { readOverflowSettings, type OverflowSettings } from "./settings.ts";
+import { getStash } from "./src/stash.ts";
+import { buildView, collectViewFromFile, collectViewFromText } from "./src/view.ts";
+import { cleanStore, entryId, storeEntry } from "./src/store.ts";
+import { runOverflowQuery } from "./src/query.ts";
+
+function formatBytes(bytes: number): string {
+	if (bytes < 1024) return `${bytes} B`;
+	if (bytes < 1_048_576) return `${(bytes / 1024).toFixed(1)} KB`;
+	return `${(bytes / 1_048_576).toFixed(1)} MB`;
+}
+
+export interface OverflowExtensionOptions {
+	settings?: OverflowSettings;
+	storeRoot?: string;
+}
+
+/**
+ * Bound stage: loaded LAST among content extensions (before system-prompt).
+ * By the time this handler runs, every other tool_result handler — RTK
+ * included — has already transformed the result, so the bounded view is
+ * built from the final visible text while the stored original comes from
+ * capture.ts's stash or bash's fullOutputPath file.
+ */
+export default function overflowExtension(pi: ExtensionAPI, options: OverflowExtensionOptions = {}): void {
+	const settings = options.settings ?? readOverflowSettings();
+	const storeRoot = () => options.storeRoot ?? join(getAgentDir(), "my-pi", "pi-overflow");
+
+	pi.on("session_start", () => {
+		// Retention/size cleanup never throws (cleanStore swallows failures)
+		// and the returned promise lets tests await it.
+		return cleanStore(storeRoot(), settings.retentionDays, settings.maxStoreBytes);
+	});
+
+	pi.on("tool_result", async (event, ctx) => {
+		const stash = getStash();
+		const staged = stash.get(event.toolCallId);
+		try {
+			if (
+				!settings.enabled ||
+				settings.exemptTools.includes(event.toolName) ||
+				event.parentToolCallId
+			) {
+				return undefined;
+			}
+			const current = event.content
+				.filter((part): part is { type: "text"; text: string } => part.type === "text")
+				.map((part) => part.text)
+				.join("\n");
+			const fullOutputPath =
+				typeof event.details?.fullOutputPath === "string"
+					? event.details.fullOutputPath
+					: staged?.fullOutputPath;
+
+			// Resolve the original: bash's spilled file wins (it is complete even
+			// when the in-context text was already cut), then the capture stash,
+			// then the current text itself.
+			let source: { kind: "text"; text: string } | { kind: "file"; path: string };
+			let originalSize: number;
+			if (fullOutputPath) {
+				try {
+					const info = await stat(fullOutputPath);
+					if (!info.isFile()) throw new Error("not a file");
+					source = { kind: "file", path: fullOutputPath };
+					originalSize = info.size;
+				} catch {
+					const fallback = staged?.rawText ?? current;
+					source = { kind: "text", text: fallback };
+					originalSize = fallback.length;
+				}
+			} else {
+				const fallback = staged?.rawText ?? current;
+				source = { kind: "text", text: fallback };
+				originalSize = fallback.length;
+			}
+
+			if (current.length <= settings.thresholdChars && originalSize <= settings.thresholdChars) {
+				return undefined;
+			}
+
+			const sessionId = ctx.sessionManager?.getSessionId?.() ?? "session";
+			const id = entryId(sessionId, event.toolCallId);
+			const stored = await storeEntry(
+				storeRoot(),
+				id,
+				source,
+				{ tool: event.toolName, args: event.input },
+				settings.maxEntryBytes,
+			);
+			const overflow = {
+				id,
+				totalLines: stored.totalLines,
+				totalBytes: stored.totalBytes,
+				truncated: current.length > settings.thresholdChars,
+			};
+
+			const images = event.content.filter((part) => part.type !== "text");
+			let text: string;
+			if (current.length > settings.thresholdChars) {
+				// The view always describes the stored file — its line numbers and
+				// totals must match what overflow_query reads, which is not true
+				// of `current` when RTK rewrote or bash pre-truncated the result.
+				const viewInput =
+					stored.storedText !== undefined
+						? collectViewFromText(stored.storedText, settings)
+						: await collectViewFromFile(join(storeRoot(), `${id}.txt`), settings);
+				text = buildView(viewInput, settings, { id, tool: event.toolName });
+			} else {
+				text = `${current}\n[pi-overflow id=${id}: full original output stored (${stored.totalLines} lines, ${formatBytes(stored.totalBytes)}); use overflow_query to inspect it]`;
+			}
+
+			return {
+				content: [{ type: "text", text }, ...images],
+				structuredContent: event.structuredContent,
+				details: { ...event.details, overflow },
+			};
+		} finally {
+			stash.delete(event.toolCallId);
+		}
+	});
+
+	pi.registerTool({
+		name: "overflow_query",
+		label: "Overflow Query",
+		description:
+			"Inspect the full original of a tool output that pi-overflow stored because it was too long to show " +
+			"(search with regex, read line ranges, or extract a JSON path).",
+		promptSnippet: "Inspect a stored oversized tool output by id: regex search, line ranges, or JSON path.",
+		promptGuidelines: [
+			"When a tool result carries a [pi-overflow id=…] marker, the model is seeing a bounded view; the complete original is stored on disk.",
+			"Call overflow_query with the marker's id and a regex, line range, or jsonPath instead of re-running the tool or asking for the full dump.",
+		],
+		parameters: Type.Object({
+			id: Type.String({ description: "12-hex storage id from a [pi-overflow id=…] marker." }),
+			search: Type.Optional(
+				Type.String({ description: "Case-insensitive regular expression matched per line." }),
+			),
+			context: Type.Optional(
+				Type.Number({ minimum: 0, maximum: 10, description: "Context lines around each search hit (default 2)." }),
+			),
+			lines: Type.Optional(
+				Type.Object({
+					start: Type.Number({ minimum: 1, description: "First line to return (1-based, inclusive)." }),
+					end: Type.Number({ minimum: 1, description: "Last line to return (inclusive)." }),
+				}),
+			),
+			jsonPath: Type.Optional(
+				Type.String({ description: "Dot/bracket path into the stored JSON, e.g. snapshot.nodes[3].name." }),
+			),
+			maxResults: Type.Optional(
+				Type.Number({ minimum: 1, maximum: 200, description: "Maximum search hits to report (default 50)." }),
+			),
+		}),
+		async execute(_toolCallId, params) {
+			const result = await runOverflowQuery(storeRoot(), params, settings);
+			return {
+				content: [{ type: "text", text: result.text }],
+				details: { error: result.error === true, truncated: result.truncated },
+				isError: result.error === true,
+			};
+		},
+	});
+}

pi-overflow/package.json

@@ -0,0 +1,49 @@
+{
+	"name": "pi-overflow",
+	"version": "0.1.0",
+	"description": "Bounded views for oversized tool results: store the full original, show head/signals/tail, recover with overflow_query.",
+	"private": true,
+	"type": "module",
+	"license": "MIT",
+	"main": "./index.ts",
+	"exports": {
+		".": "./index.ts"
+	},
+	"files": [
+		"index.ts",
+		"capture.ts",
+		"settings.ts",
+		"src",
+		"README.md",
+		"LICENSE"
+	],
+	"pi": {
+		"extensions": [
+			"./capture.ts",
+			"./index.ts"
+		]
+	},
+	"scripts": {
+		"test": "node --test test/*.test.ts",
+		"check": "node --experimental-strip-types --check index.ts && node --test test/*.test.ts"
+	},
+	"peerDependencies": {
+		"@earendil-works/pi-coding-agent": "*",
+		"typebox": "*"
+	},
+	"peerDependenciesMeta": {
+		"@earendil-works/pi-coding-agent": {
+			"optional": true
+		},
+		"typebox": {
+			"optional": true
+		}
+	},
+	"devDependencies": {
+		"@earendil-works/pi-coding-agent": "latest",
+		"typebox": "latest"
+	},
+	"engines": {
+		"node": ">=22.19.0"
+	}
+}

pi-overflow/settings.ts

@@ -0,0 +1,100 @@
+import { readFileSync } from "node:fs";
+import { fileURLToPath } from "node:url";
+
+export interface OverflowSettings {
+	enabled: boolean;
+	thresholdChars: number;
+	headLines: number;
+	tailLines: number;
+	maxSignalLines: number;
+	signalPattern: string;
+	signalRegex: RegExp;
+	maxLineChars: number;
+	exemptTools: string[];
+	retentionDays: number;
+	maxStoreBytes: number;
+	maxEntryBytes: number;
+}
+
+export const DEFAULT_SIGNAL_PATTERN =
+	"\\b(error|errors|fail|failed|failure|warn|warning|exception|traceback|panic|fatal)\\b";
+
+export const OVERFLOW_DEFAULTS = {
+	enabled: true,
+	thresholdChars: 24_000,
+	headLines: 40,
+	tailLines: 120,
+	maxSignalLines: 40,
+	signalPattern: DEFAULT_SIGNAL_PATTERN,
+	maxLineChars: 500,
+	exemptTools: ["read", "edit", "write", "overflow_query", "ask_user_question"],
+	retentionDays: 7,
+	maxStoreBytes: 1_073_741_824,
+	maxEntryBytes: 67_108_864,
+} as const;
+
+// Bundle-level `config/` is the single source of truth: the extension reads
+// `config/pi-overflow.json` from inside the installed bundle (resolved
+// relative to this package-root source file), never `settings.json`.
+export const OVERFLOW_CONFIG_PATH = fileURLToPath(
+	new URL("../config/pi-overflow.json", import.meta.url),
+);
+
+function isObject(value: unknown): value is Record<string, unknown> {
+	return value !== null && typeof value === "object" && !Array.isArray(value);
+}
+
+function bool(value: unknown, fallback: boolean): boolean {
+	return typeof value === "boolean" ? value : fallback;
+}
+
+function intInRange(value: unknown, fallback: number, min: number, max: number): number {
+	return typeof value === "number" && Number.isInteger(value) && value >= min && value <= max
+		? value
+		: fallback;
+}
+
+function compileSignalPattern(value: unknown): { pattern: string; regex: RegExp } {
+	const pattern = typeof value === "string" && value.length > 0 ? value : DEFAULT_SIGNAL_PATTERN;
+	try {
+		return { pattern, regex: new RegExp(pattern, "i") };
+	} catch {
+		return { pattern: DEFAULT_SIGNAL_PATTERN, regex: new RegExp(DEFAULT_SIGNAL_PATTERN, "i") };
+	}
+}
+
+function stringList(value: unknown, fallback: readonly string[]): string[] {
+	if (!Array.isArray(value)) return [...fallback];
+	const list = value.filter((item): item is string => typeof item === "string" && item.length > 0);
+	return [...new Set(list)];
+}
+
+/**
+ * Read `config/pi-overflow.json`. Every field is validated independently and
+ * falls back to its bundle default when missing or invalid; an unreadable or
+ * non-object file yields all defaults. Tests inject an explicit path.
+ */
+export function readOverflowSettings(configPath: string = OVERFLOW_CONFIG_PATH): OverflowSettings {
+	let parsed: unknown = {};
+	try {
+		parsed = JSON.parse(readFileSync(configPath, "utf8"));
+	} catch {
+		parsed = {};
+	}
+	const config = isObject(parsed) ? parsed : {};
+	const signal = compileSignalPattern(config.signalPattern);
+	return {
+		enabled: bool(config.enabled, OVERFLOW_DEFAULTS.enabled),
+		thresholdChars: intInRange(config.thresholdChars, OVERFLOW_DEFAULTS.thresholdChars, 1, 10_000_000),
+		headLines: intInRange(config.headLines, OVERFLOW_DEFAULTS.headLines, 0, 10_000),
+		tailLines: intInRange(config.tailLines, OVERFLOW_DEFAULTS.tailLines, 0, 10_000),
+		maxSignalLines: intInRange(config.maxSignalLines, OVERFLOW_DEFAULTS.maxSignalLines, 0, 1_000),
+		signalPattern: signal.pattern,
+		signalRegex: signal.regex,
+		maxLineChars: intInRange(config.maxLineChars, OVERFLOW_DEFAULTS.maxLineChars, 10, 100_000),
+		exemptTools: stringList(config.exemptTools, OVERFLOW_DEFAULTS.exemptTools),
+		retentionDays: intInRange(config.retentionDays, OVERFLOW_DEFAULTS.retentionDays, 0, 3_650),
+		maxStoreBytes: intInRange(config.maxStoreBytes, OVERFLOW_DEFAULTS.maxStoreBytes, 1_024, 1_099_511_627_776),
+		maxEntryBytes: intInRange(config.maxEntryBytes, OVERFLOW_DEFAULTS.maxEntryBytes, 1_024, 1_099_511_627_776),
+	};
+}

pi-overflow/src/query.ts

@@ -0,0 +1,188 @@
+import { ID_PATTERN, readStoredEntry, type EntryMeta } from "./store.ts";
+import type { OverflowSettings } from "../settings.ts";
+
+export interface QueryParams {
+	id: string;
+	search?: string;
+	context?: number;
+	lines?: { start: number; end: number };
+	jsonPath?: string;
+	maxResults?: number;
+}
+
+export interface QueryResult {
+	text: string;
+	truncated: boolean;
+	error?: boolean;
+}
+
+function fail(text: string): QueryResult {
+	return { text, truncated: false, error: true };
+}
+
+function boundOutput(text: string, limit: number): QueryResult {
+	if (text.length <= limit) return { text, truncated: false };
+	const marker = `\n[output truncated at ${limit} chars; narrow the query (smaller line range, tighter regex, or jsonPath)]`;
+	const keep = Math.max(0, limit - marker.length);
+	return { text: text.slice(0, keep) + marker, truncated: true };
+}
+
+function metaLine(id: string, meta: EntryMeta | null): string {
+	if (!meta) return `[pi-overflow id=${id}]`;
+	return `[pi-overflow id=${id} tool=${meta.tool} ${meta.totalLines} lines / ${meta.totalBytes} bytes]`;
+}
+
+/** Regexes only inspect this prefix of each stored line — a model-supplied
+ * pattern must not backtrack catastrophically across a megabyte-long line. */
+const MATCH_SCAN_CHARS = 10_000;
+
+function clipLine(line: string, maxLineChars: number): string {
+	return line.length > maxLineChars ? `${line.slice(0, maxLineChars)}…` : line;
+}
+
+function evalJsonPath(root: unknown, path: string): { ok: true; value: unknown } | { ok: false; error: string } {
+	const tokens: Array<string | number> = [];
+	const pattern = /(?:^|\.)([A-Za-z_$][A-Za-z0-9_$]*)|\[\s*(\d+)\s*\]|\[\s*["']([^"']+)["']\s*\]/g;
+	let index = 0;
+	let match: RegExpExecArray | null;
+	while ((match = pattern.exec(path)) !== null) {
+		if (match.index !== index) return { ok: false, error: `cannot parse jsonPath near "${path.slice(index)}"` };
+		index = pattern.lastIndex;
+		if (match[1] !== undefined) tokens.push(match[1]);
+		else if (match[2] !== undefined) tokens.push(Number(match[2]));
+		else tokens.push(match[3] as string);
+	}
+	if (index !== path.length || tokens.length === 0) {
+		return { ok: false, error: `cannot parse jsonPath "${path}"` };
+	}
+	let value: unknown = root;
+	for (const token of tokens) {
+		if (value === null || value === undefined) {
+			return { ok: false, error: `jsonPath stops at null/undefined before "${String(token)}"` };
+		}
+		if (typeof token === "number") {
+			if (!Array.isArray(value) || token >= value.length) {
+				return { ok: false, error: `index [${token}] out of range` };
+			}
+			value = value[token];
+		} else {
+			if (typeof value !== "object" || Array.isArray(value)) {
+				return { ok: false, error: `"${token}" is not a property of this value` };
+			}
+			const record = value as Record<string, unknown>;
+			if (!(token in record)) return { ok: false, error: `property "${token}" not found` };
+			value = record[token];
+		}
+	}
+	return { ok: true, value };
+}
+
+function mergeRanges(hits: number[], context: number): Array<{ start: number; end: number }> {
+	const ranges: Array<{ start: number; end: number }> = [];
+	for (const hit of hits) {
+		const start = Math.max(1, hit - context);
+		const end = hit + context;
+		const last = ranges[ranges.length - 1];
+		if (last && start <= last.end + 1) {
+			last.end = Math.max(last.end, end);
+		} else {
+			ranges.push({ start, end });
+		}
+	}
+	return ranges;
+}
+
+/**
+ * Execute an overflow_query against the store. Never throws on bad input —
+ * invalid ids, regexes, JSON, and paths all come back as clear error text.
+ */
+export async function runOverflowQuery(
+	root: string,
+	params: QueryParams,
+	settings: OverflowSettings,
+): Promise<QueryResult> {
+	if (typeof params.id !== "string" || !ID_PATTERN.test(params.id)) {
+		return fail(`invalid overflow id "${String(params.id)}" — expected 12 lowercase hex chars from a [pi-overflow id=…] marker`);
+	}
+	let stored: { text: string; meta: EntryMeta | null };
+	try {
+		stored = await readStoredEntry(root, params.id);
+	} catch {
+		return fail(`no stored output for id ${params.id} (it may have been cleaned up or never stored)`);
+	}
+	const lines = stored.text.split("\n");
+	const head = metaLine(params.id, stored.meta);
+	const context = Math.min(Math.max(Math.floor(params.context ?? 2), 0), 10);
+	const maxResults = Math.min(Math.max(Math.floor(params.maxResults ?? 50), 1), 200);
+
+	if (typeof params.search === "string") {
+		let regex: RegExp;
+		try {
+			regex = new RegExp(params.search, "i");
+		} catch (error) {
+			return fail(`invalid search regex: ${error instanceof Error ? error.message : String(error)}`);
+		}
+		const hits: number[] = [];
+		for (let index = 0; index < lines.length && hits.length < maxResults; index++) {
+			regex.lastIndex = 0;
+			if (regex.test(lines[index].slice(0, MATCH_SCAN_CHARS))) hits.push(index + 1);
+		}
+		if (hits.length === 0) {
+			return boundOutput(`${head}\nno lines match /${params.search}/i`, settings.thresholdChars);
+		}
+		const parts = [head, `${hits.length}${hits.length >= maxResults ? "+" : ""} matching lines:`];
+		for (const range of mergeRanges(hits, context)) {
+			const end = Math.min(range.end, lines.length);
+			parts.push(`--- lines ${range.start}-${end} ---`);
+			for (let line = range.start; line <= end; line++) {
+				parts.push(`L${line}: ${clipLine(lines[line - 1], settings.maxLineChars)}`);
+			}
+		}
+		return boundOutput(parts.join("\n"), settings.thresholdChars);
+	}
+
+	if (params.lines && typeof params.lines === "object") {
+		const start = Math.floor(params.lines.start);
+		const end = Math.floor(params.lines.end);
+		if (!Number.isFinite(start) || !Number.isFinite(end) || start < 1 || end < start) {
+			return fail(`invalid line range {start: ${params.lines.start}, end: ${params.lines.end}} — need 1 <= start <= end`);
+		}
+		const clippedEnd = Math.min(end, lines.length);
+		if (start > lines.length) {
+			return fail(`start ${start} is beyond the stored ${lines.length} lines`);
+		}
+		const parts = [head, `--- lines ${start}-${clippedEnd} ---`];
+		for (let line = start; line <= clippedEnd; line++) {
+			parts.push(`L${line}: ${clipLine(lines[line - 1], settings.maxLineChars)}`);
+		}
+		return boundOutput(parts.join("\n"), settings.thresholdChars);
+	}
+
+	if (typeof params.jsonPath === "string") {
+		let parsed: unknown;
+		try {
+			parsed = JSON.parse(stored.text);
+		} catch (error) {
+			return fail(`stored output is not valid JSON: ${error instanceof Error ? error.message : String(error)}`);
+		}
+		const result = evalJsonPath(parsed, params.jsonPath);
+		if (!result.ok) return fail(`jsonPath error: ${result.error}`);
+		let rendered: string;
+		try {
+			rendered = JSON.stringify(result.value, null, 2) ?? "undefined";
+		} catch {
+			rendered = String(result.value);
+		}
+		return boundOutput(`${head}\njsonPath ${params.jsonPath}:\n${rendered}`, settings.thresholdChars);
+	}
+
+	// No selector: metadata plus a bounded head/tail preview.
+	const headCount = Math.min(settings.headLines, lines.length);
+	const tailCount = Math.min(settings.tailLines, Math.max(0, lines.length - headCount));
+	const parts = [head, `--- lines 1-${headCount} ---`, ...lines.slice(0, headCount)];
+	if (tailCount > 0) {
+		parts.push(`… ${lines.length - headCount - tailCount} lines omitted …`);
+		parts.push(`--- lines ${lines.length - tailCount + 1}-${lines.length} ---`, ...lines.slice(lines.length - tailCount));
+	}
+	return boundOutput(parts.join("\n"), settings.thresholdChars);
+}

pi-overflow/src/stash.ts

@@ -0,0 +1,35 @@
+/**
+ * Cross-extension staging for pre-bound originals. `capture.ts` (loaded
+ * first) stashes the raw tool_result text; `index.ts` (loaded last) consumes
+ * it after every other handler — RTK in particular — has run. Sharing goes
+ * through `globalThis[Symbol.for(...)]`, the same mechanism pi-compaction
+ * uses for its metrics service.
+ */
+
+export interface StashedResult {
+	rawText: string;
+	fullOutputPath?: string;
+}
+
+const STASH_KEY = Symbol.for("my-pi.pi-overflow.stash.v1");
+const STASH_LIMIT = 64;
+
+export function getStash(): Map<string, StashedResult> {
+	const scope = globalThis as Record<symbol, unknown>;
+	let stash = scope[STASH_KEY] as Map<string, StashedResult> | undefined;
+	if (!(stash instanceof Map)) {
+		stash = new Map<string, StashedResult>();
+		scope[STASH_KEY] = stash;
+	}
+	return stash;
+}
+
+export function stashResult(stash: Map<string, StashedResult>, toolCallId: string, entry: StashedResult): void {
+	stash.delete(toolCallId);
+	stash.set(toolCallId, entry);
+	while (stash.size > STASH_LIMIT) {
+		const oldest = stash.keys().next().value;
+		if (oldest === undefined) break;
+		stash.delete(oldest);
+	}
+}

pi-overflow/src/store.ts

@@ -0,0 +1,267 @@
+import { createHash } from "node:crypto";
+import { createReadStream, createWriteStream, type WriteStream } from "node:fs";
+import { chmod, mkdir, open, readFile, readdir, rename, stat, unlink, writeFile } from "node:fs/promises";
+import { join } from "node:path";
+import { pipeline } from "node:stream/promises";
+
+export interface EntryMeta {
+	tool: string;
+	argsSummary: string;
+	totalLines: number;
+	totalBytes: number;
+	createdAt: string;
+	omittedMiddle?: boolean;
+}
+
+export interface StoredEntry {
+	id: string;
+	totalLines: number;
+	totalBytes: number;
+	/** Present when the source was text: the exact content written to disk. */
+	storedText?: string;
+}
+
+export const ID_PATTERN = /^[0-9a-f]{12}$/;
+
+const TMP_MAX_AGE_MS = 3_600_000;
+
+export function entryId(sessionId: string, toolCallId: string): string {
+	return createHash("sha1").update(`${sessionId}\0${toolCallId}`).digest("hex").slice(0, 12);
+}
+
+async function ensureDir(dir: string): Promise<void> {
+	await mkdir(dir, { recursive: true, mode: 0o700 });
+	try {
+		await chmod(dir, 0o700);
+	} catch {
+		// Best-effort permission tightening only.
+	}
+}
+
+function tmpPath(path: string): string {
+	return `${path}.tmp-${process.pid}-${Math.random().toString(36).slice(2)}`;
+}
+
+async function writeFileAtomic(path: string, content: string | Buffer): Promise<void> {
+	const tmp = tmpPath(path);
+	await writeFile(tmp, content, { mode: 0o600 });
+	await rename(tmp, path);
+}
+
+function countNewlines(text: string): number {
+	let count = 0;
+	for (let index = 0; index < text.length; index++) {
+		if (text.charCodeAt(index) === 10) count++;
+	}
+	return count;
+}
+
+export function lineCount(text: string): number {
+	if (text.length === 0) return 0;
+	return countNewlines(text) + (text.endsWith("\n") ? 0 : 1);
+}
+
+function truncateSummary(value: string, maxChars = 300): string {
+	return value.length > maxChars ? `${value.slice(0, maxChars)}…` : value;
+}
+
+export function argsSummary(input: unknown): string {
+	try {
+		return truncateSummary(JSON.stringify(input) ?? "undefined");
+	} catch {
+		return "[unserializable]";
+	}
+}
+
+/**
+ * Copy `sourcePath` into `<root>/<id>.txt` without holding it all in memory.
+ * Files larger than `maxEntryBytes` keep the first and last halves with an
+ * omission marker between them.
+ */
+async function copyFileBounded(
+	sourcePath: string,
+	destPath: string,
+	maxEntryBytes: number,
+): Promise<{ bytes: number; omittedMiddle: boolean }> {
+	const info = await stat(sourcePath);
+	if (!info.isFile()) throw new Error("full output path is not a file");
+	const size = info.size;
+	if (size <= maxEntryBytes) {
+		const tmp = tmpPath(destPath);
+		const out: WriteStream = createWriteStream(tmp, { mode: 0o600 });
+		await pipeline(createReadStream(sourcePath), out);
+		await rename(tmp, destPath);
+		return { bytes: size, omittedMiddle: false };
+	}
+	const half = Math.floor(maxEntryBytes / 2);
+	const omitted = size - half * 2;
+	const source = await open(sourcePath, "r");
+	try {
+		const head = Buffer.alloc(half);
+		const tail = Buffer.alloc(half);
+		await source.read(head, 0, half, 0);
+		await source.read(tail, 0, half, size - half);
+		const marker = Buffer.from(`\n… ${omitted} bytes omitted (exceeded ${maxEntryBytes}-byte store limit) …\n`);
+		await writeFileAtomic(destPath, Buffer.concat([head, marker, tail]));
+	} finally {
+		await source.close();
+	}
+	return { bytes: size, omittedMiddle: true };
+}
+
+async function countFileLines(path: string): Promise<number> {
+	let count = 0;
+	let lastByte = -1;
+	let seen = false;
+	for await (const chunk of createReadStream(path)) {
+		seen = true;
+		const buffer = chunk as Buffer;
+		for (let index = 0; index < buffer.length; index++) {
+			if (buffer[index] === 10) count++;
+		}
+		if (buffer.length > 0) lastByte = buffer[buffer.length - 1];
+	}
+	if (!seen) return 0;
+	return count + (lastByte === 10 ? 0 : 1);
+}
+
+/**
+ * Persist an original tool output under `<root>/<id>.txt` plus `<id>.json`
+ * metadata. Directories are 0700, files 0600, writes are atomic. The caller
+ * chooses `root` (defaults come from getAgentDir() in index.ts; tests inject
+ * a temp directory).
+ */
+export async function storeEntry(
+	root: string,
+	id: string,
+	source: { kind: "text"; text: string } | { kind: "file"; path: string },
+	meta: { tool: string; args: unknown },
+	maxEntryBytes: number,
+): Promise<StoredEntry> {
+	await ensureDir(root);
+	const textPath = join(root, `${id}.txt`);
+	const metaPath = join(root, `${id}.json`);
+	let totalLines = 0;
+	let totalBytes = 0;
+	let omittedMiddle = false;
+	let storedText: string | undefined;
+	if (source.kind === "text") {
+		let text = source.text;
+		const bytes = Buffer.byteLength(text, "utf8");
+		totalBytes = bytes;
+		if (bytes > maxEntryBytes) {
+			const halfChars = Math.floor(maxEntryBytes / 2);
+			const omitted = bytes - halfChars * 2;
+			text = `${text.slice(0, halfChars)}\n… ${omitted} bytes omitted (exceeded ${maxEntryBytes}-byte store limit) …\n${text.slice(text.length - halfChars)}`;
+			omittedMiddle = true;
+		}
+		totalLines = lineCount(text);
+		storedText = text;
+		await writeFileAtomic(textPath, text);
+	} else {
+		const copied = await copyFileBounded(source.path, textPath, maxEntryBytes);
+		totalBytes = copied.bytes;
+		omittedMiddle = copied.omittedMiddle;
+		totalLines = await countFileLines(textPath).catch(() => 0);
+	}
+	const entry: EntryMeta = {
+		tool: meta.tool,
+		argsSummary: argsSummary(meta.args),
+		totalLines,
+		totalBytes,
+		createdAt: new Date().toISOString(),
+		...(omittedMiddle ? { omittedMiddle: true } : {}),
+	};
+	await writeFileAtomic(metaPath, `${JSON.stringify(entry, null, 2)}\n`);
+	return { id, totalLines, totalBytes, ...(storedText !== undefined ? { storedText } : {}) };
+}
+
+export async function readStoredEntry(root: string, id: string): Promise<{ text: string; meta: EntryMeta | null }> {
+	if (!ID_PATTERN.test(id)) throw new Error(`invalid overflow id: ${id}`);
+	const textPath = join(root, `${id}.txt`);
+	const metaPath = join(root, `${id}.json`);
+	const text = await readFile(textPath, "utf8");
+	let meta: EntryMeta | null = null;
+	try {
+		const parsed = JSON.parse(await readFile(metaPath, "utf8"));
+		if (parsed && typeof parsed === "object") meta = parsed as EntryMeta;
+	} catch {
+		meta = null;
+	}
+	return { text, meta };
+}
+
+/**
+ * Delete tmp files left by interrupted writes (older than one hour), entries
+ * older than `retentionDays`, then evict oldest-first until the store is
+ * under `maxStoreBytes`. All failures are swallowed — cleanup must never
+ * break a session.
+ */
+export async function cleanStore(root: string, retentionDays: number, maxStoreBytes: number): Promise<void> {
+	try {
+		const now = Date.now();
+		const cutoff = now - retentionDays * 86_400_000;
+		const names = await readdir(root);
+		// Stale tmp files from interrupted writes never become entries; remove
+		// them once they are clearly abandoned.
+		for (const name of names) {
+			if (!name.includes(".tmp-")) continue;
+			try {
+				const info = await stat(join(root, name));
+				if (info.isFile() && info.mtimeMs < now - TMP_MAX_AGE_MS) {
+					await unlink(join(root, name));
+				}
+			} catch {
+				// Already gone or unreadable — nothing to do.
+			}
+		}
+		const ids = new Set(
+			names
+				.filter((name) => !name.includes(".tmp-"))
+				.map((name) => name.replace(/\.(txt|json)$/, "")),
+		);
+		const entries: Array<{ id: string; bytes: number; mtime: number }> = [];
+		for (const id of ids) {
+			let bytes = 0;
+			let mtime = 0;
+			let seen = false;
+			for (const suffix of ["txt", "json"] as const) {
+				const file = join(root, `${id}.${suffix}`);
+				try {
+					const info = await stat(file);
+					bytes += info.size;
+					mtime = Math.max(mtime, info.mtimeMs);
+					seen = true;
+				} catch {
+					// Missing half of a pair still counts toward cleanup.
+				}
+			}
+			if (seen) entries.push({ id, bytes, mtime });
+		}
+		const remove = async (id: string) => {
+			for (const suffix of ["txt", "json"] as const) {
+				await unlink(join(root, `${id}.${suffix}`)).catch(() => {});
+			}
+		};
+		let total = 0;
+		const survivors: typeof entries = [];
+		for (const entry of entries) {
+			if (entry.mtime < cutoff) {
+				await remove(entry.id);
+			} else {
+				total += entry.bytes;
+				survivors.push(entry);
+			}
+		}
+		if (total > maxStoreBytes) {
+			survivors.sort((a, b) => a.mtime - b.mtime);
+			for (const entry of survivors) {
+				if (total <= maxStoreBytes) break;
+				await remove(entry.id);
+				total -= entry.bytes;
+			}
+		}
+	} catch {
+		// Cleanup failures never propagate.
+	}
+}

pi-overflow/src/view.ts

@@ -0,0 +1,246 @@
+import { createReadStream } from "node:fs";
+import { open, stat } from "node:fs/promises";
+import type { OverflowSettings } from "../settings.ts";
+
+export interface ViewMeta {
+	id: string;
+	tool: string;
+}
+
+export interface SignalHit {
+	line: number;
+	text: string;
+}
+
+/**
+ * Pre-collected view segments of the STORED original (never the current —
+ * possibly RTK- or bash-truncated — text). Line numbers here are the stored
+ * file's line numbers, which is exactly what overflow_query reads.
+ */
+export interface ViewInput {
+	head: string[];
+	signals: SignalHit[];
+	tail: string[];
+	totalLines: number;
+	totalBytes: number;
+	/** First ~thresholdChars chars of the stored content, for char fallback. */
+	rawHead: string;
+	/** Last ~thresholdChars chars of the stored content, for char fallback. */
+	rawTail: string;
+}
+
+/** Regexes only inspect this prefix of each line — catastrophic backtracking
+ * on a giant single-line blob is not allowed to stall the handler. */
+export const MATCH_SCAN_CHARS = 10_000;
+
+function clip(line: string, maxLineChars: number): string {
+	return line.length > maxLineChars ? `${line.slice(0, maxLineChars)}…` : line;
+}
+
+/** Slice a text original into the same view segments a file scan produces. */
+export function collectViewFromText(text: string, settings: OverflowSettings): ViewInput {
+	const lines = text.split("\n");
+	const total = lines.length;
+	const headCount = Math.min(settings.headLines, total);
+	const tailCount = Math.min(settings.tailLines, Math.max(0, total - headCount));
+	const head = lines.slice(0, headCount).map((line) => clip(line, settings.maxLineChars));
+	const tail = tailCount > 0 ? lines.slice(total - tailCount).map((line) => clip(line, settings.maxLineChars)) : [];
+	const signals: SignalHit[] = [];
+	const regex = new RegExp(settings.signalRegex.source, "i");
+	for (
+		let index = headCount;
+		index < total - tailCount && signals.length < settings.maxSignalLines;
+		index++
+	) {
+		regex.lastIndex = 0;
+		if (regex.test(lines[index].slice(0, MATCH_SCAN_CHARS))) {
+			signals.push({ line: index + 1, text: clip(lines[index], settings.maxLineChars) });
+		}
+	}
+	return {
+		head,
+		signals,
+		tail,
+		totalLines: total,
+		totalBytes: Buffer.byteLength(text, "utf8"),
+		rawHead: text.slice(0, settings.thresholdChars),
+		rawTail: text.slice(Math.max(0, text.length - settings.thresholdChars)),
+	};
+}
+
+/**
+ * Stream the stored file line by line — never materialize it. Collects the
+ * first headLines lines, keeps the last tailLines lines in a ring buffer,
+ * and records at most maxSignalLines signal hits (line number + clipped
+ * text); hits landing inside the final tail range are dropped afterwards.
+ */
+export async function collectViewFromFile(path: string, settings: OverflowSettings): Promise<ViewInput> {
+	const head: string[] = [];
+	const tail: string[] = [];
+	const signals: SignalHit[] = [];
+	let totalLines = 0;
+	let remainder = "";
+	const regex = new RegExp(settings.signalRegex.source, "i");
+
+	const handleLine = (line: string) => {
+		totalLines++;
+		if (head.length < settings.headLines) {
+			head.push(clip(line, settings.maxLineChars));
+		} else if (signals.length < settings.maxSignalLines) {
+			regex.lastIndex = 0;
+			if (regex.test(line.slice(0, MATCH_SCAN_CHARS))) {
+				signals.push({ line: totalLines, text: clip(line, settings.maxLineChars) });
+			}
+		}
+		tail.push(clip(line, settings.maxLineChars));
+		if (tail.length > settings.tailLines) tail.shift();
+	};
+
+	for await (const chunk of createReadStream(path, { encoding: "utf8" })) {
+		remainder += chunk;
+		let index = remainder.indexOf("\n");
+		while (index >= 0) {
+			handleLine(remainder.slice(0, index));
+			remainder = remainder.slice(index + 1);
+			index = remainder.indexOf("\n");
+		}
+	}
+	if (remainder.length > 0) handleLine(remainder);
+
+	const tailStart = totalLines - tail.length + 1;
+	const middleSignals = signals.filter((hit) => hit.line < tailStart);
+
+	const info = await stat(path);
+	const take = Math.min(settings.thresholdChars, info.size);
+	let rawHead = "";
+	let rawTail = "";
+	if (take > 0) {
+		const handle = await open(path, "r");
+		try {
+			const headBuffer = Buffer.alloc(take);
+			const tailBuffer = Buffer.alloc(take);
+			await handle.read(headBuffer, 0, take, 0);
+			await handle.read(tailBuffer, 0, take, info.size - take);
+			rawHead = headBuffer.toString("utf8");
+			rawTail = tailBuffer.toString("utf8");
+		} finally {
+			await handle.close();
+		}
+	}
+
+	return {
+		head,
+		signals: middleSignals,
+		tail,
+		totalLines,
+		totalBytes: info.size,
+		rawHead,
+		rawTail,
+	};
+}
+
+function formatBytes(bytes: number): string {
+	if (bytes < 1024) return `${bytes} B`;
+	if (bytes < 1_048_576) return `${(bytes / 1024).toFixed(1)} KB`;
+	if (bytes < 1_073_741_824) return `${(bytes / 1_048_576).toFixed(1)} MB`;
+	return `${(bytes / 1_073_741_824).toFixed(2)} GB`;
+}
+
+function signalDescription(pattern: string): string {
+	const compact = pattern.replace(/\\b/g, "").replace(/[()]/g, "");
+	return compact.length > 60 ? `${compact.slice(0, 60)}…` : compact;
+}
+
+/**
+ * Deterministic bounded view of the stored original: head lines, signal
+ * lines from the middle section, and tail lines — all with the stored
+ * file's line numbers. Output never exceeds settings.thresholdChars and the
+ * tail of the stored content is always visible.
+ *
+ * Shrink order under budget pressure is tail > signals > head: head first
+ * collapses to min(10, headLines), signals halve stepwise to 10 then 0, then
+ * head/tail shrink ~35/65 with tail preferred, and char truncation of the
+ * stored head/tail is the final fallback.
+ */
+export function buildView(input: ViewInput, settings: OverflowSettings, meta: ViewMeta): string {
+	const budget = settings.thresholdChars;
+	const header =
+		`[pi-overflow id=${meta.id} tool=${meta.tool} ${input.totalLines} lines / ${formatBytes(input.totalBytes)}` +
+		` — full output stored; use overflow_query to search or read ranges]`;
+
+	const render = (headLines: string[], signals: SignalHit[], tailLines: string[]): string => {
+		const parts: string[] = [header];
+		if (headLines.length > 0) {
+			parts.push(`--- lines 1-${headLines.length} ---`, ...headLines);
+		}
+		if (signals.length > 0) {
+			parts.push(
+				`--- signals (lines matching ${signalDescription(settings.signalPattern)}, up to ${settings.maxSignalLines}) ---`,
+				...signals.map((hit) => `L${hit.line}: ${hit.text}`),
+			);
+		}
+		if (tailLines.length > 0) {
+			const first = input.totalLines - tailLines.length + 1;
+			parts.push(`--- lines ${first}-${input.totalLines} ---`, ...tailLines);
+		}
+		return parts.join("\n");
+	};
+
+	const charFallback = (): string => {
+		const markerReserve = 64;
+		const bodyBudget = Math.max(64, budget - header.length - markerReserve);
+		const headChars = Math.floor(bodyBudget * 0.35);
+		const tailChars = bodyBudget - headChars;
+		const headText = input.rawHead.slice(0, headChars);
+		const tailText = input.rawTail.slice(Math.max(0, input.rawTail.length - tailChars));
+		const omitted = Math.max(0, input.totalBytes - headText.length - tailText.length);
+		return `${header}\n${headText}\n… ${omitted} chars omitted …\n${tailText}`;
+	};
+
+	let headList = input.head;
+	let signalList = input.signals;
+	let tailList = input.tail;
+
+	let result = render(headList, signalList, tailList);
+	if (result.length <= budget) return result;
+
+	// 1. Collapse head to its minimum useful window.
+	headList = headList.slice(0, Math.min(10, settings.headLines));
+	result = render(headList, signalList, tailList);
+	if (result.length <= budget) return result;
+
+	// 2. Halve signals stepwise until 10 remain, then drop them entirely.
+	while (signalList.length > 10) {
+		signalList = signalList.slice(0, Math.ceil(signalList.length / 2));
+		result = render(headList, signalList, tailList);
+		if (result.length <= budget) return result;
+	}
+	if (signalList.length > 0) {
+		signalList = [];
+		result = render(headList, signalList, tailList);
+		if (result.length <= budget) return result;
+	}
+
+	// When head+tail already span every stored line there is no line-level
+	// slack left: go straight to char truncation so the ending survives.
+	if (input.head.length + input.tail.length >= input.totalLines) return charFallback();
+
+	// 3. Shrink head/tail ~35/65; tail keeps the larger share and always keeps
+	// its last line (the stored ending).
+	while (headList.length + tailList.length > 0) {
+		result = render(headList, signalList, tailList);
+		if (result.length <= budget) return result;
+		const headChars = headList.reduce((sum, line) => sum + line.length + 1, 0);
+		const tailChars = tailList.reduce((sum, line) => sum + line.length + 1, 0);
+		if (tailList.length > 1 && (tailChars >= headChars * (65 / 35) || headList.length === 0)) {
+			tailList.shift();
+		} else if (headList.length > 0) {
+			headList.pop();
+		} else {
+			tailList.shift();
+		}
+	}
+
+	// 4. Last resort: keep the stored text's own head 35% / tail 65%.
+	return charFallback();
+}

pi-overflow/test/handler.test.ts

@@ -0,0 +1,314 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { mkdtemp, readFile, readdir, stat, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import overflowCapture from "../capture.ts";
+import overflowExtension from "../index.ts";
+import { getStash } from "../src/stash.ts";
+import { readOverflowSettings, OVERFLOW_DEFAULTS, type OverflowSettings } from "../settings.ts";
+
+function harness() {
+	const handlers = new Map<string, Function[]>();
+	const tools = new Map<string, any>();
+	const pi: any = {
+		on(name: string, handler: Function) {
+			const list = handlers.get(name) ?? [];
+			list.push(handler);
+			handlers.set(name, list);
+		},
+		registerTool(tool: any) { tools.set(tool.name, tool); },
+	};
+	return { pi, handlers, tools };
+}
+
+function ctx() {
+	return { sessionManager: { getSessionId: () => "session-test" } } as any;
+}
+
+function settings(overrides: Partial<OverflowSettings> = {}): OverflowSettings {
+	return { ...readOverflowSettings("/nonexistent/pi-overflow.json"), ...overrides };
+}
+
+function textEvent(toolCallId: string, text: string, extra: Record<string, unknown> = {}) {
+	return {
+		toolCallId,
+		toolName: "bash",
+		input: { command: "run" },
+		content: [{ type: "text", text }],
+		details: {},
+		...extra,
+	};
+}
+
+async function runHandlers(handlers: Map<string, Function[]>, event: any, context: any) {
+	let current = event;
+	for (const handler of handlers.get("tool_result") ?? []) {
+		const result = await handler(current, context);
+		if (result) current = { ...current, ...result };
+	}
+	return current;
+}
+
+async function makeStore() {
+	return mkdtemp(join(tmpdir(), "pi-overflow-test-"));
+}
+
+test("capture stashes the original before another handler shortens it; bound stage stores original but views current", async () => {
+	const root = await makeStore();
+	getStash().clear();
+	const capture = harness();
+	overflowCapture(capture.pi);
+	const bound = harness();
+	overflowExtension(bound.pi, { storeRoot: root, settings: settings() });
+
+	const big = Array.from({ length: 3000 }, (_, i) => `orig ${i}`).join("\n");
+	const event = textEvent("call-1", big);
+	const afterCapture = await runHandlers(capture.handlers, event, ctx());
+	assert.equal(afterCapture, event, "capture must not modify the result");
+	assert.ok(getStash().has("call-1"), "capture must stash the original");
+
+	// Simulate RTK: replace content with a shortened version that is still
+	// over threshold.
+	const rtkShortened = Array.from({ length: 900 }, (_, i) => `rtk ${i}`).join("\n") + "\n" + "z".repeat(30_000);
+	const mid = { ...afterCapture, content: [{ type: "text", text: rtkShortened }] };
+	const final = await runHandlers(bound.handlers, mid, ctx());
+
+	assert.ok(!getStash().has("call-1"), "stash entry must be consumed");
+	const text = final.content[0].text;
+	assert.match(text, /\[pi-overflow id=[0-9a-f]{12} tool=bash/);
+	assert.ok(text.length <= settings().thresholdChars);
+
+	const id = final.details.overflow.id;
+	const stored = await readFile(join(root, `${id}.txt`), "utf8");
+	assert.ok(stored.includes("orig 0"), "store must keep the ORIGINAL, not the RTK-shortened text");
+	assert.ok(stored.includes("orig 2999"));
+	assert.equal(final.details.overflow.truncated, true);
+});
+
+test("details.fullOutputPath file is used as the original", async () => {
+	const root = await makeStore();
+	getStash().clear();
+	const bound = harness();
+	overflowExtension(bound.pi, { storeRoot: root, settings: settings() });
+
+	const spill = join(root, "spill.txt");
+	await writeFile(spill, Array.from({ length: 5000 }, (_, i) => `spill ${i}`).join("\n"));
+	const event = textEvent("call-2", "x".repeat(30_000), { details: { fullOutputPath: spill } });
+	const final = await runHandlers(bound.handlers, event, ctx());
+	const stored = await readFile(join(root, `${final.details.overflow.id}.txt`), "utf8");
+	assert.ok(stored.includes("spill 0"));
+	assert.ok(stored.includes("spill 4999"));
+});
+
+test("view line numbers track the STORED file, not the current text", async () => {
+	const root = await makeStore();
+	getStash().clear();
+	const bound = harness();
+	overflowExtension(bound.pi, { storeRoot: root, settings: settings() });
+
+	// bash spill: 5000 lines on disk, but the in-context result only carries
+	// the last ~100 lines (still over threshold via padding).
+	const spill = join(root, "spill-5000.txt");
+	await writeFile(
+		spill,
+		Array.from({ length: 5000 }, (_, i) => `spill line ${i + 1}`).join("\n"),
+	);
+	const current = Array.from({ length: 100 }, (_, i) => `spill line ${4901 + i}`).join("\n") + "\n" + "pad".repeat(10_000);
+	const event = textEvent("call-lines", current, { details: { fullOutputPath: spill } });
+	const final = await runHandlers(bound.handlers, event, ctx());
+
+	const text = final.content[0].text;
+	const id = final.details.overflow.id;
+	assert.match(text, /--- lines 4881-5000 ---/, "tail range must come from the stored 5000-line file");
+	assert.match(text, /spill line 5000/);
+	assert.match(text, /\[pi-overflow id=[0-9a-f]{12} tool=bash 5000 lines/);
+
+	// overflow_query with the view's tail range returns the same lines.
+	const tool = bound.tools.get("overflow_query");
+	const result = await tool.execute("q1", { id, lines: { start: 4881, end: 4883 } });
+	assert.match(result.content[0].text, /L4881: spill line 4881/);
+	assert.match(result.content[0].text, /L4883: spill line 4883/);
+});
+
+test("RTK scenario: view line numbers come from the stored original text", async () => {
+	const root = await makeStore();
+	getStash().clear();
+	const capture = harness();
+	overflowCapture(capture.pi);
+	const bound = harness();
+	overflowExtension(bound.pi, { storeRoot: root, settings: settings() });
+
+	const original = Array.from({ length: 3000 }, (_, i) => `orig ${i + 1}`).join("\n");
+	await runHandlers(capture.handlers, textEvent("call-rtk", original), ctx());
+	// RTK rewrites the visible text but it stays over threshold.
+	const mid = {
+		...textEvent("call-rtk", ""),
+		content: [{ type: "text", text: "z".repeat(30_000) }],
+	};
+	const final = await runHandlers(bound.handlers, mid, ctx());
+	const text = final.content[0].text;
+	assert.match(text, /--- lines 2881-3000 ---/, "view tail must be the stored original's tail");
+	assert.ok(text.includes("orig 3000"));
+});
+
+test("file sources over maxEntryBytes keep real head and tail halves", async () => {
+	const root = await makeStore();
+	const { storeEntry } = await import("../src/store.ts");
+	const src = join(root, "big.bin");
+	const content = "HEAD-".repeat(200) + "MID-".repeat(2000) + "TAIL-".repeat(200);
+	await writeFile(src, content);
+	const maxEntryBytes = 2000;
+	await storeEntry(root, "0123456789ab", { kind: "file", path: src }, { tool: "bash", args: {} }, maxEntryBytes);
+	const stored = await readFile(join(root, "0123456789ab.txt"), "utf8");
+	assert.ok(stored.startsWith("HEAD-"));
+	assert.ok(stored.endsWith("TAIL-"));
+	assert.match(stored, /bytes omitted/);
+	// Head and tail halves must equal the source's real bytes.
+	const half = Math.floor(maxEntryBytes / 2);
+	assert.equal(stored.slice(0, half), content.slice(0, half));
+	assert.equal(stored.slice(-half), content.slice(-half));
+	assert.ok(!stored.includes("\0".repeat(100)), "must not store zeroed content");
+});
+
+test("interrupted-write tmp files older than an hour are cleaned", async () => {
+	const root = await makeStore();
+	const bound = harness();
+	overflowExtension(bound.pi, { storeRoot: root, settings: settings() });
+
+	const staleTmp = join(root, "deadbeef.txt.tmp-999-zzz");
+	const freshTmp = join(root, "livebeef.txt.tmp-111-aaa");
+	await writeFile(staleTmp, "partial");
+	await writeFile(freshTmp, "partial");
+	const old = new Date(Date.now() - 2 * 3_600_000);
+	const { utimes } = await import("node:fs/promises");
+	await utimes(staleTmp, old, old);
+
+	for (const handler of bound.handlers.get("session_start") ?? []) await handler({}, ctx());
+	const { existsSync } = await import("node:fs");
+	assert.ok(!existsSync(staleTmp), "stale tmp must be removed");
+	assert.ok(existsSync(freshTmp), "fresh tmp must survive");
+});
+
+test("current under threshold but original over → one-line marker appended", async () => {
+	const root = await makeStore();
+	getStash().clear();
+	const capture = harness();
+	overflowCapture(capture.pi);
+	const bound = harness();
+	overflowExtension(bound.pi, { storeRoot: root, settings: settings() });
+
+	const big = "a".repeat(30_000);
+	const event = textEvent("call-3", big);
+	await runHandlers(capture.handlers, event, ctx());
+	const mid = { ...event, content: [{ type: "text", text: "short summary" }] };
+	const final = await runHandlers(bound.handlers, mid, ctx());
+
+	const text = final.content[0].text;
+	assert.ok(text.startsWith("short summary"));
+	assert.match(text, /\[pi-overflow id=[0-9a-f]{12}: full original output stored \(\d+ lines, .*; use overflow_query/);
+	assert.equal(final.details.overflow.truncated, false);
+	assert.ok(final.details.overflow.id);
+});
+
+test("exempt tools and nested calls are ignored", async () => {
+	const root = await makeStore();
+	getStash().clear();
+	const bound = harness();
+	overflowExtension(bound.pi, { storeRoot: root, settings: settings() });
+
+	const big = "x".repeat(30_000);
+	const exempt = await runHandlers(bound.handlers, textEvent("c-ex", big, { toolName: "read" }), ctx());
+	assert.equal(exempt.content[0].text, big, "read results must pass through untouched");
+
+	const nested = await runHandlers(bound.handlers, textEvent("c-nest", big, { parentToolCallId: "parent" }), ctx());
+	assert.equal(nested.content[0].text, big, "nested calls must pass through untouched");
+});
+
+test("image parts and structuredContent survive", async () => {
+	const root = await makeStore();
+	getStash().clear();
+	const bound = harness();
+	overflowExtension(bound.pi, { storeRoot: root, settings: settings() });
+
+	const event = textEvent("call-4", "x".repeat(30_000));
+	event.content.push({ type: "image", data: "b64" } as any);
+	(event as any).structuredContent = { rows: 3 };
+	const final = await runHandlers(bound.handlers, event, ctx());
+	assert.deepEqual(final.structuredContent, { rows: 3 });
+	assert.equal(final.content.at(-1).type, "image");
+	assert.equal(final.content.at(-1).data, "b64");
+});
+
+test("existing details fields survive the merge", async () => {
+	const root = await makeStore();
+	getStash().clear();
+	const bound = harness();
+	overflowExtension(bound.pi, { storeRoot: root, settings: settings() });
+
+	const event = textEvent("call-5", "x".repeat(30_000), { details: { rtk: { saved: 100 }, fullOutputPath: undefined } });
+	const final = await runHandlers(bound.handlers, event, ctx());
+	assert.deepEqual(final.details.rtk, { saved: 100 });
+	assert.ok(final.details.overflow.id);
+});
+
+test("files are 0600 and directories 0700", async () => {
+	const root = await makeStore();
+	getStash().clear();
+	const bound = harness();
+	overflowExtension(bound.pi, { storeRoot: root, settings: settings() });
+
+	const final = await runHandlers(bound.handlers, textEvent("call-6", "x".repeat(30_000)), ctx());
+	const id = final.details.overflow.id;
+	const dirMode = (await stat(root)).mode & 0o777;
+	const fileMode = (await stat(join(root, `${id}.txt`))).mode & 0o777;
+	const metaMode = (await stat(join(root, `${id}.json`))).mode & 0o777;
+	assert.equal(dirMode, 0o700);
+	assert.equal(fileMode, 0o600);
+	assert.equal(metaMode, 0o600);
+});
+
+test("retention and total-size cleanup remove old entries", async () => {
+	const root = await makeStore();
+	const bound = harness();
+	overflowExtension(bound.pi, {
+		storeRoot: root,
+		settings: settings({ retentionDays: 1, maxStoreBytes: 100 }),
+	});
+
+	// An old entry (mtime in the past) must be deleted by age.
+	const oldId = "aaaaaaaaaaaa";
+	await writeFile(join(root, `${oldId}.txt`), "old");
+	await writeFile(join(root, `${oldId}.json`), "{}");
+	const past = new Date(Date.now() - 10 * 86_400_000);
+	const { utimes } = await import("node:fs/promises");
+	await utimes(join(root, `${oldId}.txt`), past, past);
+	await utimes(join(root, `${oldId}.json`), past, past);
+
+	for (const handler of bound.handlers.get("session_start") ?? []) await handler({}, ctx());
+	const names = await readdir(root);
+	assert.ok(!names.some((n) => n.startsWith(oldId)), "old entry must be deleted");
+
+	// Fresh entries exceeding maxStoreBytes get evicted oldest-first.
+	for (const [i, id] of ["bbbbbbbbbbbb", "cccccccccccc", "dddddddddddd"].entries()) {
+		await writeFile(join(root, `${id}.txt`), "y".repeat(80));
+		const t = new Date(Date.now() - (3 - i) * 60_000);
+		await utimes(join(root, `${id}.txt`), t, t);
+	}
+	for (const handler of bound.handlers.get("session_start") ?? []) await handler({}, ctx());
+	const after = (await readdir(root)).filter((n) => n.endsWith(".txt"));
+	assert.ok(after.length <= 2, `expected eviction under maxStoreBytes, got ${after.join(",")}`);
+	assert.ok(!after.includes("bbbbbbbbbbbb.txt"), "oldest entry must be evicted first");
+});
+
+test("disabled setting clears the stash and returns undefined", async () => {
+	const root = await makeStore();
+	getStash().clear();
+	const bound = harness();
+	overflowExtension(bound.pi, { storeRoot: root, settings: settings({ enabled: false }) });
+	getStash().set("c-off", { rawText: "x".repeat(30_000) });
+	const event = textEvent("c-off", "x".repeat(30_000));
+	const result = await runHandlers(bound.handlers, event, ctx());
+	assert.equal(result, event);
+	assert.ok(!getStash().has("c-off"));
+});

pi-overflow/test/query.test.ts

@@ -0,0 +1,112 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { mkdtemp, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { runOverflowQuery } from "../src/query.ts";
+import { storeEntry } from "../src/store.ts";
+import { readOverflowSettings, type OverflowSettings } from "../settings.ts";
+
+function settings(overrides: Partial<OverflowSettings> = {}): OverflowSettings {
+	return { ...readOverflowSettings("/nonexistent/pi-overflow.json"), ...overrides };
+}
+
+async function store(text: string, id = "0123456789ab") {
+	const root = await mkdtemp(join(tmpdir(), "pi-overflow-query-"));
+	await storeEntry(root, id, { kind: "text", text }, { tool: "bash", args: { command: "run" } }, 1_000_000);
+	return root;
+}
+
+test("search returns hits with merged context", async () => {
+	const lines = Array.from({ length: 100 }, (_, i) => (i === 9 || i === 11 || i === 50 ? `line ${i + 1} FAIL` : `line ${i + 1}`));
+	const root = await store(lines.join("\n"));
+	const result = await runOverflowQuery(root, { id: "0123456789ab", search: "FAIL", context: 1 }, settings());
+	assert.ok(!result.error);
+	assert.match(result.text, /L10: line 10 FAIL/);
+	assert.match(result.text, /L12: line 12 FAIL/);
+	assert.match(result.text, /L51: line 51 FAIL/);
+	// hits at lines 10 and 12 with context 1 merge into one range 9-13
+	assert.match(result.text, /--- lines 9-13 ---/);
+	assert.match(result.text, /--- lines 50-52 ---/);
+});
+
+test("line-range retrieval returns numbered lines", async () => {
+	const lines = Array.from({ length: 50 }, (_, i) => `row ${i + 1}`);
+	const root = await store(lines.join("\n"));
+	const result = await runOverflowQuery(root, { id: "0123456789ab", lines: { start: 10, end: 15 } }, settings());
+	assert.match(result.text, /--- lines 10-15 ---/);
+	assert.match(result.text, /L10: row 10/);
+	assert.match(result.text, /L15: row 15/);
+	assert.ok(!result.text.includes("row 16"));
+});
+
+test("jsonPath extracts nested values", async () => {
+	const doc = JSON.stringify({ snapshot: { nodes: [{ name: "a" }, { name: "b" }, { name: "c" }] } });
+	const root = await store(doc);
+	const result = await runOverflowQuery(root, { id: "0123456789ab", jsonPath: "snapshot.nodes[2].name" }, settings());
+	assert.match(result.text, /"c"/);
+});
+
+test("invalid id, regex, JSON and path return clear errors without throwing", async () => {
+	const root = await store('{"a": {"b": 1}}');
+	assert.ok((await runOverflowQuery(root, { id: "../etc/passwd" }, settings())).error);
+	assert.ok((await runOverflowQuery(root, { id: "zzzzzzzzzzzz" }, settings())).error);
+	assert.ok((await runOverflowQuery(root, { id: "0123456789ab", search: "([" }, settings())).error);
+	const goodPath = await runOverflowQuery(root, { id: "0123456789ab", jsonPath: "a.b" }, settings());
+	assert.ok(!goodPath.error, "valid path on valid JSON must succeed");
+	const badPath = await runOverflowQuery(root, { id: "0123456789ab", jsonPath: "a.b.c!!" }, settings());
+	assert.ok(badPath.error);
+	const missing = await runOverflowQuery(root, { id: "ffffffffffff" }, settings());
+	assert.ok(missing.error);
+});
+
+test("non-JSON content + jsonPath reports the parse failure", async () => {
+	const root = await store("not json at all");
+	const result = await runOverflowQuery(root, { id: "0123456789ab", jsonPath: "a" }, settings());
+	assert.ok(result.error);
+	assert.match(result.text, /not valid JSON/);
+});
+
+test("output is bounded by thresholdChars and marked truncated", async () => {
+	const lines = Array.from({ length: 5000 }, (_, i) => `line ${i + 1} pad ${"p".repeat(40)}`);
+	const root = await store(lines.join("\n"));
+	const result = await runOverflowQuery(
+		root,
+		{ id: "0123456789ab", lines: { start: 1, end: 5000 } },
+		settings({ thresholdChars: 4000 }),
+	);
+	assert.ok(result.text.length <= 4000);
+	assert.equal(result.truncated, true);
+	assert.match(result.text, /truncated/);
+});
+
+test("lines and search output clip each line to maxLineChars", async () => {
+	const long = `row ${"q".repeat(2_000)}`;
+	const root = await store(`first\n${long}\n${long} ERROR\nlast`);
+	const s = settings({ maxLineChars: 60 });
+	const lines = await runOverflowQuery(root, { id: "0123456789ab", lines: { start: 2, end: 3 } }, s);
+	assert.ok(lines.text.includes("q".repeat(56) + "…"));
+	assert.ok(!lines.text.includes("q".repeat(61)));
+	const search = await runOverflowQuery(root, { id: "0123456789ab", search: "ERROR", context: 0 }, s);
+	assert.ok(search.text.includes("q".repeat(56) + "…"));
+	assert.ok(!search.text.includes("q".repeat(61)));
+});
+
+test("search only scans the first 10000 chars of each line", async () => {
+	const earlyHit = `hit ${"a".repeat(100)} FAIL`;
+	const lateHit = `${"a".repeat(10_500)} FAIL`;
+	const root = await store(`${earlyHit}\n${lateHit}\nplain`);
+	const result = await runOverflowQuery(root, { id: "0123456789ab", search: "FAIL", context: 0 }, settings());
+	assert.match(result.text, /L1: hit /);
+	assert.ok(!result.text.includes("L2:"), "match beyond char 10000 must not be reported");
+});
+
+test("default query returns metadata and a first-screen view", async () => {
+	const lines = Array.from({ length: 300 }, (_, i) => `row ${i + 1}`);
+	const root = await store(lines.join("\n"));
+	const result = await runOverflowQuery(root, { id: "0123456789ab" }, settings());
+	assert.ok(!result.error);
+	assert.match(result.text, /\[pi-overflow id=0123456789ab tool=bash \d+ lines \/ \d+ bytes\]/);
+	assert.match(result.text, /--- lines 1-40 ---/);
+	assert.ok(result.text.includes("row 300"), "tail of the stored file must appear");
+});

pi-overflow/test/view.test.ts

@@ -0,0 +1,86 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { buildView, collectViewFromText } from "../src/view.ts";
+import { readOverflowSettings, type OverflowSettings } from "../settings.ts";
+
+function settings(overrides: Partial<OverflowSettings> = {}): OverflowSettings {
+	return {
+		...readOverflowSettings("/nonexistent/pi-overflow.json"),
+		...overrides,
+	};
+}
+
+const meta = { id: "abc123def456", tool: "bash" };
+
+function view(text: string, s = settings()) {
+	return buildView(collectViewFromText(text, s), s, meta);
+}
+
+test("below threshold remains unchanged", () => {
+	const out = view("line one\nline two\nline three");
+	assert.ok(out.includes("line one"));
+	assert.ok(out.includes("line three"));
+});
+
+test("above threshold includes header, line ranges, signal lines and tail", () => {
+	const lines: string[] = [];
+	for (let i = 1; i <= 1000; i++) lines.push(`row ${i} ${i === 500 ? "ERROR boom" : "ok"}`);
+	const out = view(lines.join("\n"));
+	assert.match(out, /\[pi-overflow id=abc123def456 tool=bash 1000 lines/);
+	assert.match(out, /--- lines 1-40 ---/);
+	assert.match(out, /L500: row 500 ERROR boom/);
+	assert.match(out, /--- lines 881-1000 ---/);
+	assert.ok(out.includes("row 1000"));
+});
+
+test("output never exceeds thresholdChars", () => {
+	const lines = Array.from({ length: 5000 }, (_, i) => `line ${i} ${"x".repeat(60)}`);
+	const s = settings({ thresholdChars: 2000 });
+	const out = view(lines.join("\n"), s);
+	assert.ok(out.length <= 2000, `view ${out.length} exceeds budget`);
+	assert.ok(out.includes("line 4999"), "tail must survive");
+});
+
+test("few long lines still expose the ending", () => {
+	const lines = ["start", ...Array.from({ length: 5 }, (_, i) => `middle${i} ${"y".repeat(4000)}`), "THE_END_MARKER"];
+	const s = settings({ thresholdChars: 3000, headLines: 40, tailLines: 120 });
+	const out = view(lines.join("\n"), s);
+	assert.ok(out.includes("THE_END_MARKER"));
+	assert.ok(out.length <= 3000 + 100, `view ${out.length} far exceeds budget`);
+});
+
+test("same input produces identical output", () => {
+	const lines = Array.from({ length: 800 }, (_, i) => `entry ${i} ${i % 77 === 0 ? "warning" : "fine"}`);
+	const text = lines.join("\n");
+	assert.equal(view(text), view(text));
+});
+
+test("a single line over maxLineChars is truncated", () => {
+	const long = "z".repeat(2000);
+	const out = view(`a\n${long}\nb`, settings({ maxLineChars: 50 }));
+	assert.ok(out.includes("z".repeat(50) + "…"));
+	assert.ok(!out.includes("z".repeat(51)));
+});
+
+test("signals outrank head and survive when head+tail exceed the budget", () => {
+	// ~150-char lines: head 40 + tail 120 = 160 lines * 150 = ~24K plus
+	// overhead, over budget. Shrink order must keep some signals while head
+	// collapses to 10 lines before the 35/65 phase kicks in.
+	const lines = Array.from({ length: 400 }, (_, i) => {
+		const marker = i === 200 ? "ERROR detected here" : "ok";
+		return `line ${i + 1} ${marker} ${"x".repeat(140)}`;
+	});
+	const out = view(lines.join("\n"));
+	assert.match(out, /--- signals/);
+	assert.match(out, /L201: line 201 ERROR detected here/);
+	// head collapsed below the configured 40 lines
+	assert.match(out, /--- lines 1-(?:[0-9]|10) ---/);
+	assert.ok(out.includes("line 400"));
+});
+
+test("signal hits inside the tail region are not duplicated", () => {
+	const lines = Array.from({ length: 100 }, (_, i) => `line ${i + 1} ${"y".repeat(40)}`);
+	lines[99] = "line 100 tail-side ERROR";
+	const out = view(lines.join("\n"));
+	assert.ok(!out.includes("L100:"), "tail-region hit must not appear in signals");
+});

pi-permission-system/AGENTS.md

@@ -8,7 +8,7 @@
 - reviewer 默认跟随会话当前主模型（每次审查实时读取 `ctx.model`），经 `modelRegistry.streamSimple(...).result()` 调用；`autoReview.provider` / `autoReview.model` 必须同时设置（固定审查模型）或同时留空，只设其一视为配置错误并 `defer`。默认 `reasoning: "low"`（仅对支持推理的模型生效）、总超时 30 秒、最多 2 次尝试（仅 Provider 错误重试）、`maxOutputTokens` 4000。只把 active branch 中的直接用户消息与已识别、未取消的结构化问答作为授权证据，assistant/tool/compaction 内容不能自行授权；工具结果可替换为更短的 pi-compaction 摘要，但仍属不可信证据。同一条用户消息内完全相同的请求只缓存 `approve` 结论。
 - `config/pi-permission-policy.jsonc` 必须显式配置 `authorizerChain: ["auto-review"]`，普通未匹配请求和需要自动复核的类别声明为 `ask`；普通 pattern 必须按具体度优先（精确匹配、字面字符更多、通配符更少、pattern 更长，最后才以声明顺序破同级）；`autoReview` 段只保留 `enabled`，不设 provider/model 种子值。`config/pi-permission-hard-guards.jsonc` 独立保存可由用户调整但不能被普通策略、Session grant、per-agent 规则或 reviewer 覆盖的高危 `deny`，以及 `protectedWrites` 人工写保护清单。
 - `protectedWrites` 是确定性的 human-only 写路径，不属于普通策略也不是第二个授权范围：命中目标的 `write`/`edit` 与本地 `bash` 写入（redirect-out、写操作程序参数、`bash -c`/`eval` 等包装载荷中的受保护路径）一律弹出人工确认，跳过 AutoReview、yolo 与既有 Session grant，弹窗不提供「本会话允许」选项；没有可用确认界面（无 UI、子代理转发终端）时直接拒绝。读取不受影响；`ssh_bash` 等远端 shell 工具不参与本机路径保护。
-- 策略默认允许常规只读与低风险工具；`web_fetch`、`ctx_fetch_and_index`、`skill_manage` 保持 `ask`（模型自选的 URL 可能外传数据，skill 写入会进入所有后续会话）；硬防御默认只保留确定性的破坏性 Shell 命令；Git 只读管道命令放行，其余非只读操作、包管理、文件/系统/网络高风险操作与 `mcp__<server>__<tool>` 调用按唯一全局 policy 判定。项目外路径不产生独立 `external_directory` 判断，只走配置中的工具、`path` 与 Bash pattern。
+- 策略默认允许常规只读与低风险工具；`web_fetch`、`skill_manage` 保持 `ask`（模型自选的 URL 可能外传数据，skill 写入会进入所有后续会话）；硬防御默认只保留确定性的破坏性 Shell 命令；Git 只读管道命令放行，其余非只读操作、包管理、文件/系统/网络高风险操作与 `mcp__<server>__<tool>` 调用按唯一全局 policy 判定。项目外路径不产生独立 `external_directory` 判断，只走配置中的工具、`path` 与 Bash pattern。
 - `HardGuardPolicyLoader` 必须保持 global-only、无 SessionRules、无项目覆盖；唯一的组合包 hard-guards 文件不可用时 fail closed 到 deny-all，并报告配置问题。普通 policy 与 hard guard 配置不得合并成一份文件。
 - 内置 AutoReview 实现位于 `src/auto-review/`；同步原 AutoReview 上游时必须从源码仓库明确 tag/commit 移植，保留 `LICENSES/pi-permission-auto-review-MIT.txt`、作者和 `docs/auto-review/UPSTREAM.md`，不得恢复顶层独立扩展、npm 预编译入口或平行 gate。
 - reviewer 只能处理普通策略已判定为 `ask` 且未被 hard guard 拒绝的请求；不得绕过硬 `deny`。

pi-rtk-optimizer/AGENTS.md

@@ -2,7 +2,7 @@
 
 - 扩展入口是 `pi-rtk-optimizer/index.ts`，主要实现位于 `pi-rtk-optimizer/src/`。
 - 组合包运行环境为 Node.js 22.19 或更高版本；本子目录开发验证还需要 npm、Bun 和项目声明的开发依赖。
-- RTK 默认只压缩非搜索输出，包括本地 `bash` 与远端 `ssh_bash` 的 ANSI 清理、测试聚合、构建过滤、Git 压缩、Lint 聚合和兜底截断；`ssh_bash` 只复用输出处理，远端命令不得进入 RTK rewrite。
+- RTK 默认只压缩非搜索输出，包括本地 `bash` 与远端 `ssh_bash` 的 ANSI 清理、测试聚合、构建过滤、Git 压缩、Lint 聚合；兜底截断默认关闭，新结果的展示截断由 pi-overflow 统一负责；`ssh_bash` 只复用输出处理，远端命令不得进入 RTK rewrite。
 - RTK 命令改写默认关闭（`commandRewritingEnabled` 默认 `false`）；默认安装不依赖系统 `rtk` CLI，只有用户主动开启命令改写时才需要 `rtk rewrite`，其命令支持策略由已安装的 `rtk rewrite` 决定，不在扩展内重复维护规则。
 - FFF override 独占搜索工具与搜索输出；RTK 不得处理本地或远端搜索工具的调用/结果（含 `grep` 工具结果），也不得通过默认命令改写接管 `rg`、`grep`、`find`、`fd`。
 - `readCompaction.enabled`、`sourceCodeFilteringEnabled` 和 `smartTruncate.enabled` 当前均默认 `false`，源码读取保持原样；未经用户明确决定，不因节省上下文而改变这些默认值。若以后开启 read 压缩，优先考虑 `readCompaction + smartTruncate`，源码过滤仍独立评估；必须保留精确 `offset/limit` 读取、短文件和行锚点的完整性。

pi-rtk-optimizer/README.md

@@ -154,7 +154,7 @@ When `read` output uses Pi hashline/anchor prefixes, the compactor treats each a
 |--------|------|---------|-------|-------------|
 | `outputCompaction.smartTruncate.enabled` | boolean | `false` | — | Enable smart line-based truncation for read output when read compaction is enabled |
 | `outputCompaction.smartTruncate.maxLines` | number | `220` | 40–4000 | Maximum lines after smart truncation |
-| `outputCompaction.truncate.enabled` | boolean | `true` | — | Enable hard character truncation |
+| `outputCompaction.truncate.enabled` | boolean | `false` | — | Enable hard character truncation (off in this bundle — pi-overflow owns display truncation) |
 | `outputCompaction.truncate.maxChars` | number | `12000` | 1000–200000 | Maximum characters in final output |
 
 ### Source Code Filtering Levels
@@ -195,7 +195,7 @@ When `read` output uses Pi hashline/anchor prefixes, the compactor treats each a
       "maxLines": 220
     },
     "truncate": {
-      "enabled": true,
+      "enabled": false,
       "maxChars": 12000
     }
   }

pi-rtk-optimizer/UPSTREAM.md

@@ -13,3 +13,4 @@
 
 - 默认只负责非搜索工具输出压缩；RTK 命令改写收敛为可选能力（`commandRewritingEnabled` 默认 `false`）。
 - `readCompaction`、`sourceCodeFiltering`、`smartTruncate` 默认关闭，源码读取保持原样。
+- `outputCompaction.truncate` 兜底截断默认关闭（组合包 `config/rtk-optimizer.json`），新结果的展示截断由 pi-overflow 统一负责；RTK 只保留语义压缩。

pi-tool-search/AGENTS.md

@@ -3,7 +3,7 @@
 - 扩展入口是 `pi-tool-search/extensions/index.ts`，组合包权威分组位于 `extensions/bundle-groups.ts`，目录生成/校验/缓存位于 `extensions/catalog.ts`，配置读取与路径解析位于 `extensions/config.ts`（直接读取组合包 `config/tool-search.json`；目录缓存仍在 `<agentDir>/my-pi/tool-search/catalog-v1.json`），测试保留在 `test/`。
 - 同步上游必须记录明确 tag/commit，保留 README、CHANGELOG、文档、作者和许可证说明；不得导入上游 `.pi`、嵌套 `.git`、`node_modules` 或构建产物。
 - 根包必须通过 `file:./pi-tool-search` 提供本地依赖并直接加载仓库源码，不得同时安装或加载 npm `pi-tool-search` 实现；同步上游时先核对本地增量动态工具改造，禁止用旧 npm 入口或上游旧宿主实现直接覆盖。
-- 默认常驻核心 `read` / `write` / `edit` / `bash` / `grep` / `find`、`codegraph_explore`、`lsp_diagnostics`、`ctx_execute` / `ctx_execute_file` / `ctx_batch_execute`、六个记忆工具与 `context_tree_query`，以及 `tool_search`；Windows 上内置 `powershell` 必须与 `bash` 同属固定本地 Shell 核心工具，非 Windows 平台必须从目录中过滤。配置从已安装组合包内的 `config/tool-search.json` 直接读取，文件缺失或无效时回退 `BUNDLE_TOOL_SEARCH_DEFAULTS`；扩展自身不播种、不迁移，也不再读写 `settings.json` 的 `toolSearch` 段。
+- 默认常驻核心 `read` / `write` / `edit` / `bash` / `grep` / `find`、`codegraph_explore`、`lsp_diagnostics`、六个记忆工具与 `context_tree_query`、`overflow_query`，以及 `tool_search`；Windows 上内置 `powershell` 必须与 `bash` 同属固定本地 Shell 核心工具，非 Windows 平台必须从目录中过滤。配置从已安装组合包内的 `config/tool-search.json` 直接读取，文件缺失或无效时回退 `BUNDLE_TOOL_SEARCH_DEFAULTS`；扩展自身不播种、不迁移，也不再读写 `settings.json` 的 `toolSearch` 段。
 - 其余组合包工具必须优先使用 `extensions/bundle-groups.ts` 中提交的权威工作流分组；精确已有分组必须直接激活，不得等待无关未知工具的模型增强。默认最多 5 个动态组、模型组最多 8 个工具、动态工具总数最多 28；`groups` 可原子激活最多 5 个精确组，完整请求超限时不得改变当前状态；加载或组内调用更新 LRU，超限先移除最久未用的非请求组。自然语言查询必须优先精确工具名和已提交 workflow alias，弱匹配或近似同分只能返回候选，不得任意激活；`/tool-search-status` 必须显示当前容量、LRU 顺序与最近淘汰原因。
 - 标准组合包工具全部命中预置目录时不得调用模型、创建用户缓存或发送隐藏 schema。只有存在未识别第三方工具时才允许惰性模型增强；生成结果必须保留全部预置 assignment，模型元数据不得替代真实名称、参数或 schema。缓存必须位于 agent 目录且权限 `600`；模型失败必须继续使用预置目录 + 确定性未知工具分组。`groupOverrides` 优先级高于预置分组。
 - 必须直接使用当前 `@earendil-works/pi-coding-agent`、`typebox`、`ModelRegistry.complete()` 和 `setActiveTools()`；首次纯增加载使用当前 Pi 的增量结果传播，LRU 替换允许走宿主安全 fallback。不得安装或映射旧 `@mariozechner` runtime，不得恢复 provider payload 改写、代理分发或隐藏 `sendMessage` steer/retry 兼容循环。Tool Search 是唯一可调用 `getActiveTools()` / `setActiveTools()` 管理工具显隐的组合包扩展；`pi-permission-system` 只能在调用时裁决。

pi-tool-search/CHANGELOG.md

@@ -3,6 +3,7 @@
 ## Unreleased — my-pi maintained source
 
 ### Changed
+- Removed the retired context-mode integration: the `ctx_execute` / `ctx_execute_file` / `ctx_batch_execute` pins left `alwaysEnabled` and the four `context-*` bundle groups were dropped. Large tool results are now bounded by pi-overflow and recoverable through the pinned `overflow_query`.
 - Imported upstream `v0.3.6` at `ddfb23646fd3957b791214de278e23aa393c9b13` for direct local maintenance.
 - Migrated runtime imports to `@earendil-works/pi-coding-agent` and current `typebox`.
 - Replaced the hidden steer/retry compatibility loop with current Pi dynamic-tool propagation.
@@ -12,6 +13,7 @@
 - Added checked-in authoritative groups for every tool exposed by the standard my-pi bundle, eliminating first-use model generation and user caching unless unrecognized third-party tools are present.
 - Replaced the Hermes and third-party Hippo assignments with the five tools exposed by the official Hippo Pi extension, grouped into recall and management workflows.
 - Replaced the retired Hippo `hippo_*` assignments with the vendored Hermes memory extension's six tools, grouped into `memory-recall` (`memory_search`, `session_search`, `context_tree_query`) and `memory-management` (`memory_add`, `memory_replace`, `memory_remove`, `skill_manage`).
+- Added `overflow_query` to `memory-recall` and pinned it in `alwaysEnabled` — inspecting a pi-overflow-stored original must not depend on a `tool_search` activation.
 - Added the enabled bundle's `Agent`, `get_subagent_result`, and `steer_subagent` tools to a checked-in `subagents` orchestration group.
 - Added authoritative `ssh-connection`, `ssh-remote-files`, `ssh-remote-search`, and `ssh-remote-shell` groups for the maintained `pi-ssh` tools, keeping reviewed connection, bounded discovery, structured file operations, and reviewed shell execution separate; the connection workflow now requires sequential `ssh_connect` to succeed before dependent tools.
 - Added sequential `ssh_cd` to the authoritative remote-shell workflow and direct the model to wait for a successful workspace change before dependent remote calls, avoiding both the unknown-tools fallback and cwd races.

pi-tool-search/README.md

@@ -33,10 +33,6 @@ Run `/tool-search-status` to inspect the current catalog source, pinned count, d
 | `ssh-remote-shell` | `ssh_cd`, `ssh_bash` |
 | `code-intelligence` | CodeGraph and all LSP tools |
 | `web-search` | `web_search`, `web_fetch` (pi-web-search unified tools) |
-| `context-execution` | `ctx_execute`, `ctx_execute_file`, `ctx_batch_execute` |
-| `context-knowledge` | Context indexing, searching, and web-doc ingestion |
-| `context-observability` | Context stats, doctor, and Insight |
-| `context-administration` | Context upgrade and destructive purge |
 | `memory-recall` | Hermes memory/session search and condensed-output recovery |
 | `memory-management` | Hermes memory writes and procedural skill management |
 | `subagents` | Launch agents, collect background results, and steer running agents |
@@ -45,7 +41,7 @@ Run `/tool-search-status` to inspect the current catalog source, pinned count, d
 | `chrome-interaction` | Chrome click/type/fill/key/hover/drag/tap/scroll input |
 | `chrome-debugging` | Chrome page evaluation, console/network diagnostics, response retrieval, and file upload |
 
-Pinned tools are removed from these dynamic groups at runtime. The defaults pin `codegraph_explore`, `lsp_diagnostics`, the three high-frequency Context Mode execution tools, and all six Hermes memory tools, so none of them consume dynamic group or tool capacity.
+Pinned tools are removed from these dynamic groups at runtime. The defaults pin `codegraph_explore`, `lsp_diagnostics`, and all six Hermes memory tools, so none of them consume dynamic group or tool capacity.
 
 ## Loading and eviction
 
@@ -78,16 +74,14 @@ Pi reads this configuration directly from the installed bundle's `config/tool-se
   "alwaysEnabled": [
     "codegraph_explore",
     "lsp_diagnostics",
-    "ctx_execute",
-    "ctx_execute_file",
-    "ctx_batch_execute",
     "memory_search",
     "session_search",
     "memory_add",
     "memory_replace",
     "memory_remove",
     "skill_manage",
-    "context_tree_query"
+    "context_tree_query",
+    "overflow_query"
   ],
   "showToolSearchFooterStatus": false,
   "maxActiveGroups": 5,
@@ -99,7 +93,7 @@ Pi reads this configuration directly from the installed bundle's `config/tool-se
 
 | Key | Bundle default | Description |
 |---|---|---|
-| `alwaysEnabled` | CodeGraph explore, LSP diagnostics, Context execution trio, all six Hermes memory tools, `context_tree_query` | Exact names pinned alongside Pi core tools |
+| `alwaysEnabled` | CodeGraph explore, LSP diagnostics, all six Hermes memory tools, `context_tree_query`, `overflow_query` | Exact names pinned alongside Pi core tools |
 | `showToolSearchFooterStatus` | `false` | Show active tool/group counts |
 | `maxActiveGroups` | `5` | Maximum simultaneously active dynamic groups |
 | `maxToolsPerGroup` | `8` | Maximum size accepted from model-generated groups |

pi-tool-search/docs/dynamic-tool-loading.md

@@ -2,7 +2,7 @@
 
 `pi-tool-search` keeps full low-frequency schemas hidden behind a compact, validated group catalog.
 
-1. `session_start` activates `tool_search`, the six cross-platform Pi core tools, and configured `alwaysEnabled` names. The my-pi defaults also pin CodeGraph exploration, LSP diagnostics, the three high-frequency Context Mode execution tools, and the six Hermes memory tools. On Windows, `powershell` is pinned beside `bash`; on non-Windows hosts it is excluded. Pinned tools do not consume the dynamic-group or dynamic-tool budget.
+1. `session_start` activates `tool_search`, the six cross-platform Pi core tools, and configured `alwaysEnabled` names. The my-pi defaults also pin CodeGraph exploration, LSP diagnostics, and the six Hermes memory tools. On Windows, `powershell` is pinned beside `bash`; on non-Windows hosts it is excluded. Pinned tools do not consume the dynamic-group or dynamic-tool budget.
 2. Hidden definitions are hashed with grouping constraints and matched against `extensions/bundle-groups.ts`. Available standard tools receive checked-in groups; unavailable optional tools are omitted.
 3. If every hidden tool is recognized, the bundle catalog is used directly. No model call or per-user cache is required.
 4. Additional tools receive immediate deterministic prefix/source groups. A matching model-enriched cache is reused only if it preserves all checked-in bundle assignments.
@@ -13,7 +13,7 @@
 9. Natural-language lookup gives exact tool names and workflow aliases more weight than descriptive overlap. Weak or closely tied matches return candidates without activation.
 10. `turn_start` notices tool or grouping-configuration changes by hash, resets stale dynamic groups, and reconstructs the checked-in/hybrid catalog.
 
-The standard bundle assigns the Hermes memory extension's six tools to recall and management groups: `memory_search` and `session_search` join `context_tree_query` in `memory-recall`, while `memory_add`, `memory_replace`, `memory_remove`, and `skill_manage` form `memory-management`. All six are pinned by default, because the injected `<memory-policy>` tells the model to consult and write memory during normal work, and a schema the model cannot see would silently disable that loop. Third-party `tff-memory_*` names and the retired Hippo `hippo_*` names are not seeded. It also assigns `Agent`, `get_subagent_result`, and `steer_subagent` to the `subagents` orchestration group.
+The standard bundle assigns the Hermes memory extension's six tools to recall and management groups: `memory_search`, `session_search`, and `overflow_query` join `context_tree_query` in `memory-recall`, while `memory_add`, `memory_replace`, `memory_remove`, and `skill_manage` form `memory-management`. All six are pinned by default, because the injected `<memory-policy>` tells the model to consult and write memory during normal work, and a schema the model cannot see would silently disable that loop. Third-party `tff-memory_*` names and the retired Hippo `hippo_*` names are not seeded. It also assigns `Agent`, `get_subagent_result`, and `steer_subagent` to the `subagents` orchestration group.
 The maintained `pi-ssh` tools are also split by workflow: sequential reviewed agent-controlled connection lives in `ssh-connection`, structured `ssh_read`/`ssh_write`/`ssh_edit` file operations live in `ssh-remote-files`, bounded `ssh_find`/`ssh_grep` discovery lives in `ssh-remote-search`, and sequential reviewed workspace changes plus reviewed command execution live in `ssh-remote-shell`. The workflows tell the model to call `ssh_connect` or `ssh_cd` separately and wait for success before issuing remote calls that depend on the new connection or cwd.
 The maintained `ask_user_question` tool has its own `user-interaction` group so the model can load the complete interactive schema only when a missing decision or preference must come from the user.
 The pinned `pi-chrome` bridge registers its 21 `chrome_*` tools only after explicit per-session authorization. Once registered, catalog refresh assigns them to `chrome-navigation`, `chrome-interaction`, and `chrome-debugging`; the 28-tool default allows all three groups to coexist. Revocation remains enforced inside every upstream tool execution even if a schema is later reactivated by Tool Search.

pi-tool-search/extensions/bundle-groups.ts

@@ -85,50 +85,14 @@ export const BUNDLE_GROUP_DEFINITIONS: BundleGroupDefinition[] = [
     keywords: ["web", "search", "fetch", "news", "docs", "网页", "搜索", "新闻", "文档"],
     tools: ["web_search", "web_fetch"],
   },
-  {
-    id: "context-execution",
-    title: "Context-isolated execution",
-    summary: "Run commands or analyze large files in a sandbox while returning only compact derived output.",
-    useWhen: ["Logs, tests, builds, generated data, commands, or workspace files may produce large output"],
-    avoidWhen: ["Exact source lines are needed for an anchored edit"],
-    keywords: ["execute", "large file", "logs", "tests", "build", "大文件", "日志", "测试", "构建"],
-    tools: ["ctx_execute", "ctx_execute_file", "ctx_batch_execute"],
-  },
-  {
-    id: "context-knowledge",
-    title: "Context knowledge base",
-    summary: "Index local or web documentation and retrieve focused passages from the persistent context-mode knowledge base.",
-    useWhen: ["Documentation or session knowledge should be stored and queried without rereading raw content"],
-    avoidWhen: ["A one-shot small source can be read directly"],
-    keywords: ["index", "search", "knowledge base", "fetch docs", "索引", "知识库", "文档检索"],
-    tools: ["ctx_index", "ctx_search", "ctx_fetch_and_index"],
-  },
-  {
-    id: "context-observability",
-    title: "Context observability",
-    summary: "Inspect context-mode health, savings statistics, and the hosted Insight dashboard.",
-    useWhen: ["Diagnosing context-mode or reviewing context savings and analytics"],
-    avoidWhen: ["The task is normal code or data processing"],
-    keywords: ["stats", "doctor", "insight", "health", "统计", "诊断", "上下文节省"],
-    tools: ["ctx_stats", "ctx_doctor", "ctx_insight"],
-  },
-  {
-    id: "context-administration",
-    title: "Context administration",
-    summary: "Upgrade context-mode or destructively purge indexed context data.",
-    useWhen: ["The user explicitly asks to upgrade context-mode or purge a named scope"],
-    avoidWhen: ["Routine searching, indexing, or performance inspection"],
-    keywords: ["upgrade", "purge", "delete", "升级", "清除", "删除知识库"],
-    tools: ["ctx_upgrade", "ctx_purge"],
-  },
   {
     id: "memory-recall",
     title: "Persistent memory recall",
-    summary: "Search durable user, global, project, and failure memories, search indexed past conversations, or recover pruned tool-call output.",
+    summary: "Search durable user, global, project, and failure memories, search indexed past conversations, or recover pruned or overflow-stored tool-call output.",
     useWhen: ["The task may depend on durable context from previous sessions: preferences, project conventions, prior decisions, known failures, corrections, or condensed outputs"],
     avoidWhen: ["The required evidence is already in the current visible context, or the question is a generic one-off that durable memory would not help"],
-    keywords: ["memory", "recall", "search", "session", "history", "recover", "记忆", "召回", "检索", "历史会话", "恢复输出"],
+    keywords: ["memory", "recall", "search", "session", "history", "recover", "overflow", "记忆", "召回", "检索", "历史会话", "恢复输出"],
-    tools: ["memory_search", "session_search", "context_tree_query"],
+    tools: ["memory_search", "session_search", "context_tree_query", "overflow_query"],
   },
   {
     id: "memory-management",

pi-tool-search/extensions/catalog.ts

@@ -335,7 +335,6 @@ export function writeCachedCatalog(cachePath: string, catalog: ToolCatalog): voi
 }
 
 const GROUP_QUERY_ALIASES: Record<string, string[]> = {
-  "context-execution": ["context mode", "run command", "test build logs", "执行命令", "运行测试", "分析日志"],
   "ssh-connection": ["ssh connect", "remote server connection", "连接服务器", "远程连接"],
   "ssh-remote-shell": ["ssh shell", "remote command", "remote terminal", "远程命令", "远程终端"],
   "ssh-remote-files": ["ssh file", "remote read write edit", "远程文件", "远程读写"],

pi-tool-search/extensions/config.ts

@@ -25,9 +25,6 @@ export const BUNDLE_TOOL_SEARCH_DEFAULTS = {
   alwaysEnabled: [
     "codegraph_explore",
     "lsp_diagnostics",
-    "ctx_execute",
-    "ctx_execute_file",
-    "ctx_batch_execute",
     "memory_search",
     "session_search",
     "memory_add",
@@ -35,6 +32,7 @@ export const BUNDLE_TOOL_SEARCH_DEFAULTS = {
     "memory_remove",
     "skill_manage",
     "context_tree_query",
+    "overflow_query",
   ],
   showToolSearchFooterStatus: false,
   maxActiveGroups: 5,

pi-tool-search/test/bundle-groups.test.ts

@@ -25,7 +25,7 @@ test("bundle seed names are unique and cover the declared my-pi tools", () => {
   assert.deepEqual(result.unknownTools, []);
   assert.equal(result.coveredNames.size, names.length);
   assert.equal(result.catalog.tools.length, names.length);
-  assert.equal(result.catalog.tools.find((card) => card.name === "ctx_purge")?.primaryGroup, "context-administration");
+  assert.equal(result.catalog.tools.find((card) => card.name === "overflow_query")?.primaryGroup, "memory-recall");
   assert.equal(result.catalog.tools.find((card) => card.name === "web_fetch")?.primaryGroup, "web-search");
   assert.equal(result.catalog.tools.find((card) => card.name === "memory_search")?.primaryGroup, "memory-recall");
   assert.equal(result.catalog.tools.find((card) => card.name === "session_search")?.primaryGroup, "memory-recall");

pi-tool-search/test/config.test.ts

@@ -33,9 +33,6 @@ test("bundle defaults keep the committed tool list", () => {
   assert.deepEqual(BUNDLE_TOOL_SEARCH_DEFAULTS.alwaysEnabled, [
     "codegraph_explore",
     "lsp_diagnostics",
-    "ctx_execute",
-    "ctx_execute_file",
-    "ctx_batch_execute",
     "memory_search",
     "session_search",
     "memory_add",
@@ -43,6 +40,7 @@ test("bundle defaults keep the committed tool list", () => {
     "memory_remove",
     "skill_manage",
     "context_tree_query",
+    "overflow_query",
   ]);
   assert.deepEqual(bundleFallbackConfig(), {
     alwaysEnabled: [...BUNDLE_TOOL_SEARCH_DEFAULTS.alwaysEnabled],

pi-tool-search/test/extension.test.ts

@@ -254,9 +254,9 @@ test("uses the precomputed bundle groups without calling a model", async () => {
       sourceTool("read", "Read a file"),
       sourceTool("web_search", "Search broadly for current web sources"),
       sourceTool("web_fetch", "Fetch a selected web source"),
-      sourceTool("ctx_execute", "Run commands in a context sandbox"),
+      sourceTool("context_tree_query", "Inspect the context tree ledger"),
-      sourceTool("ctx_execute_file", "Analyze a file in a context sandbox"),
+      sourceTool("overflow_query", "Inspect a stored oversized tool result"),
-      sourceTool("ctx_batch_execute", "Run a bounded command batch"),
+      sourceTool("memory_search", "Search durable memories"),
     ];
     const api = {
       getAllTools: () => sourceTools,
@@ -280,7 +280,7 @@ test("uses the precomputed bundle groups without calling a model", async () => {
       } as unknown as ExtensionContext["modelRegistry"],
     });
     handlers.get("session_start")?.({}, ctx);
-    assert.ok(["ctx_execute", "ctx_execute_file", "ctx_batch_execute"].every((name) => activeCalls.at(-1)?.includes(name)));
+    assert.ok(["context_tree_query", "overflow_query", "memory_search"].every((name) => activeCalls.at(-1)?.includes(name)));
 
     const ambiguous = await registered.get("tool_search")?.execute(
       "call-1",

pi-web-search/README.md

@@ -6,9 +6,12 @@ Unified web search and fetch for Pi. The main model only ever sees two tools —
 - `web_search(query, max_results?, site?, published_after?, published_before?, depth?)`
   routes each query to Tavily, Exa, or Keenable using the session's current main
   model (`ctx.model` + `ctx.modelRegistry.complete`), guided by per-provider
-  rules. Results include `details.provider` / `details.routedBy` for debugging.
+  rules. `max_results` defaults to 8 and is clamped to 1–20. Results include
+  `details.provider` / `details.routedBy` for debugging.
 - `web_fetch(urls, max_characters?)` always reads pages via Exa
   (`web_fetch_exa`), degrading silently to Keenable then Tavily.
+  `max_characters` defaults to 20,000 and is clamped to 1–100,000 per page;
+  oversized fetch output is further bounded by pi-overflow downstream.
 
 ## Providers (`clients/`)
 

pi-web-search/index.ts

@@ -21,8 +21,15 @@ import {
 } from "./router.ts";
 
 const DEFAULT_MAX_RESULTS = 8;
+const MAX_RESULTS_LIMIT = 20;
 const MAX_FETCH_URLS = 10;
-const DEFAULT_FETCH_MAX_CHARACTERS = 3000;
+const DEFAULT_FETCH_MAX_CHARACTERS = 20000;
+const FETCH_MAX_CHARACTERS_LIMIT = 100000;
+
+function clampInt(value: number | undefined, fallback: number, min: number, max: number): number {
+	if (typeof value !== "number" || !Number.isFinite(value)) return fallback;
+	return Math.min(Math.max(Math.floor(value), min), max);
+}
 
 export interface WebSearchParams {
 	query: string;
@@ -48,7 +55,7 @@ function formatSearchPayload(payload: unknown): string {
 export function toTavilyParams(params: WebSearchParams) {
 	return {
 		query: params.query,
-		max_results: params.max_results ?? DEFAULT_MAX_RESULTS,
+		max_results: clampInt(params.max_results, DEFAULT_MAX_RESULTS, 1, MAX_RESULTS_LIMIT),
 		search_depth: (params.depth ?? "basic") as "basic" | "advanced",
 		include_domains: params.site ? [params.site] : undefined,
 		start_date: params.published_after,
@@ -59,7 +66,7 @@ export function toTavilyParams(params: WebSearchParams) {
 export function toExaParams(params: WebSearchParams) {
 	return {
 		query: params.query,
-		numResults: params.max_results ?? DEFAULT_MAX_RESULTS,
+		numResults: clampInt(params.max_results, DEFAULT_MAX_RESULTS, 1, MAX_RESULTS_LIMIT),
 		includeDomains: params.site ? [params.site] : undefined,
 		startPublishedDate: params.published_after,
 		endPublishedDate: params.published_before,
@@ -171,7 +178,7 @@ export async function dispatchFetch(
 	if (urls.length > MAX_FETCH_URLS) {
 		return toResult(`web_fetch accepts at most ${MAX_FETCH_URLS} URLs per call; received ${urls.length}.`, { error: true });
 	}
-	const maxCharacters = params.max_characters ?? DEFAULT_FETCH_MAX_CHARACTERS;
+	const maxCharacters = clampInt(params.max_characters, DEFAULT_FETCH_MAX_CHARACTERS, 1, FETCH_MAX_CHARACTERS_LIMIT);
 	const order: SearchProvider[] = ["exa", "keenable"];
 	if (tavilyAvailable(resolve)) order.push("tavily");
 
@@ -213,7 +220,9 @@ export default function webSearchExtension(pi: ExtensionAPI): void {
 				},
 				max_results: {
 					type: "number",
-					description: `Maximum number of results to return (default ${DEFAULT_MAX_RESULTS}).`,
+					minimum: 1,
+					maximum: MAX_RESULTS_LIMIT,
+					description: `Maximum number of results to return (default ${DEFAULT_MAX_RESULTS}, max ${MAX_RESULTS_LIMIT}).`,
 				},
 				site: {
 					type: "string",
@@ -268,7 +277,9 @@ export default function webSearchExtension(pi: ExtensionAPI): void {
 				},
 				max_characters: {
 					type: "number",
-					description: `Maximum characters to extract per page (default ${DEFAULT_FETCH_MAX_CHARACTERS}).`,
+					minimum: 1,
+					maximum: FETCH_MAX_CHARACTERS_LIMIT,
+					description: `Maximum characters to extract per page (default ${DEFAULT_FETCH_MAX_CHARACTERS}, max ${FETCH_MAX_CHARACTERS_LIMIT}).`,
 				},
 			},
 			required: ["urls"],

tests/extension-load.test.ts

@@ -10,13 +10,16 @@ import test from "node:test";
 const repositoryRoot = resolve(dirname(fileURLToPath(import.meta.url)), "..");
 const hasPi = spawnSync("pi", ["--version"], { encoding: "utf8" }).status === 0;
 
+const filesIncludePackageDir = (files: string[], packageDir: string): boolean =>
+	files.some((entry) => entry === packageDir || entry.startsWith(`${packageDir}/`));
+
 test("enabled subagents wrapper loads with its bundle-owned mandatory permission entry", { skip: !hasPi }, async () => {
   const packageJson = JSON.parse(await readFile(join(repositoryRoot, "package.json"), "utf8")) as {
     dependencies: Record<string, string>;
     files: string[];
     pi: { extensions: string[] };
   };
-  assert.ok(packageJson.files.includes("pi-subagents"));
+  assert.ok(filesIncludePackageDir(packageJson.files, "pi-subagents"));
   assert.equal(packageJson.dependencies["@tintinweb/pi-subagents"], "file:./pi-subagents");
   assert.ok(
     packageJson.pi.extensions.includes("./extensions/subagents.ts"),
@@ -48,8 +51,8 @@ test("all package extensions load together without global registration conflicts
 	const packageJson = JSON.parse(await readFile(join(repositoryRoot, "package.json"), "utf8")) as {
 		dependencies: Record<string, string>;
 		files: string[];
-		pi: { extensions: string[] };
+		pi: { extensions: string[]; skills?: string[] };
-		allowScripts: Record<string, boolean>;
+		allowScripts?: Record<string, boolean>;
 	};
 	assert.ok(
 		packageJson.pi.extensions.includes("./extensions/codegraph.ts"),
@@ -74,7 +77,7 @@ test("all package extensions load together without global registration conflicts
 		"the bundle must load the locally maintained minimal footer",
 	);
 	assert.equal(packageJson.dependencies["pi-compaction"], "file:./pi-compaction");
-	assert.ok(packageJson.files.includes("pi-compaction"), "the packed bundle must include maintained pi-compaction source");
+	assert.ok(filesIncludePackageDir(packageJson.files, "pi-compaction"), "the packed bundle must include maintained pi-compaction source");
 	assert.ok(
 		packageJson.pi.extensions.includes("./pi-compaction/index.ts"),
 		"the bundle must load the unified pi-compaction authority",
@@ -97,20 +100,20 @@ test("all package extensions load together without global registration conflicts
 	assert.equal(packageJson.dependencies["pi-ssh"], "file:./pi-ssh");
 	assert.match(packageJson.dependencies.ssh2 ?? "", /^\d+\.\d+\.\d+$/, "the packed bundle must install pi-ssh's runtime transport at an exact version");
 	assert.match(packageJson.dependencies.jiti ?? "", /^\d+\.\d+\.\d+$/, "the packed helper must load local TypeScript outside Pi at an exact version");
-	assert.ok(packageJson.files.includes("pi-ssh"), "the packed bundle must include the locally maintained SSH source");
+	assert.ok(filesIncludePackageDir(packageJson.files, "pi-ssh"), "the packed bundle must include the locally maintained SSH source");
 	assert.ok(packageJson.files.includes("configure.sh"), "the packed bundle must include the search/SSH config helper");
 	assert.ok(packageJson.pi.extensions.includes("./pi-ask-user/index.ts"), "the bundle must load pi-ask-user");
 	assert.equal(packageJson.dependencies["pi-ask-user"], "file:./pi-ask-user");
-	assert.ok(packageJson.files.includes("pi-ask-user"), "the packed bundle must include pi-ask-user source");
+	assert.ok(filesIncludePackageDir(packageJson.files, "pi-ask-user"), "the packed bundle must include pi-ask-user source");
 	assert.ok(
 		packageJson.pi.extensions.includes("./pi-worktree/index.ts"),
 		"the bundle must load the locally maintained worktree commands",
 	);
 	assert.equal(packageJson.dependencies["pi-worktree"], "file:./pi-worktree");
-	assert.ok(packageJson.files.includes("pi-worktree"), "the packed bundle must include pi-worktree source");
+	assert.ok(filesIncludePackageDir(packageJson.files, "pi-worktree"), "the packed bundle must include pi-worktree source");
 	const chromeExtension = "./pi-chrome/extensions/chrome-profile-bridge/index.ts";
 	assert.equal(packageJson.dependencies["pi-chrome"], "file:./pi-chrome", "the browser bridge must use the locally maintained source");
-	assert.ok(packageJson.files.includes("pi-chrome"), "the packed bundle must include the locally maintained pi-chrome source");
+	assert.ok(filesIncludePackageDir(packageJson.files, "pi-chrome"), "the packed bundle must include the locally maintained pi-chrome source");
 	assert.ok(packageJson.pi.extensions.includes(chromeExtension), "the bundle must load the local Chrome profile bridge directly");
 	assert.ok(
 		packageJson.pi.extensions.indexOf(chromeExtension) <
@@ -167,7 +170,7 @@ test("all package extensions load together without global registration conflicts
 	);
 	assert.equal(packageJson.dependencies["@smoose/pi-notify"], "file:./pi-notify");
 	assert.ok(
-		packageJson.files.includes("pi-notify"),
+		filesIncludePackageDir(packageJson.files, "pi-notify"),
 		"the packed bundle must include the locally maintained notification source",
 	);
 	assert.equal(packageJson.dependencies["@ogulcancelik/pi-minimal-footer"], "file:./pi-minimal-footer");
@@ -195,14 +198,34 @@ test("all package extensions load together without global registration conflicts
 		"./extensions/system-prompt.ts",
 		"the single system-prompt owner must be the last loaded extension",
 	);
-	assert.ok(
+	// pi-overflow has two entry points with strict ordering: capture.ts must
-		packageJson.pi.extensions.includes("./extensions/context-mode.ts"),
+	// run first so it sees results before RTK rewrites them, and index.ts must
-		"Context Mode must load through the bundle's prompt-guarded wrapper",
+	// run last (before system-prompt) so it bounds the final visible text.
+	assert.equal(
+		packageJson.pi.extensions.at(0),
+		"./pi-overflow/capture.ts",
+		"pi-overflow capture must be the first loaded extension",
+	);
+	assert.equal(packageJson.dependencies["pi-overflow"], "file:./pi-overflow");
+	assert.ok(filesIncludePackageDir(packageJson.files, "pi-overflow"), "the packed bundle must include pi-overflow source");
+	assert.equal(
+		packageJson.pi.extensions.at(-2),
+		"./pi-overflow/index.ts",
+		"pi-overflow bound stage must load after all other content extensions and before system-prompt",
 	);
 	assert.equal(
-		packageJson.pi.extensions.some((entry) => entry.includes("context-mode/build/")),
+		packageJson.dependencies["context-mode"],
+		undefined,
+		"context-mode was retired; new tool-result bounding is owned by pi-overflow",
+	);
+	assert.equal(
+		packageJson.pi.extensions.some((entry) => entry.includes("context-mode")),
 		false,
-		"the raw Context Mode adapter must not be loaded alongside the wrapper",
+		"no context-mode entry point may be loaded",
+	);
+	assert.ok(
+		!Array.isArray(packageJson.pi.skills) || packageJson.pi.skills.length === 0,
+		"pi.skills must be absent or empty now that the context-mode skills are gone",
 	);
 	assert.equal(
 		packageJson.pi.extensions.includes("./extensions/system-prompt-cover.ts"),
@@ -227,7 +250,7 @@ test("all package extensions load together without global registration conflicts
 	}
 	assert.equal(packageJson.dependencies["pi-hermes-memory"], "file:./pi-hermes-memory");
 	assert.ok(
-		packageJson.files.includes("pi-hermes-memory"),
+		filesIncludePackageDir(packageJson.files, "pi-hermes-memory"),
 		"the packed bundle must include the locally maintained Hermes memory source",
 	);
 	assert.ok(
@@ -242,28 +265,28 @@ test("all package extensions load together without global registration conflicts
 	);
 	assert.equal(packageJson.dependencies["@the-forge-flow/hippo-memory-pi"], undefined);
 	assert.equal(packageJson.dependencies["better-sqlite3"], undefined);
-	// allowScripts must approve exactly the root lockfile's better-sqlite3
+	// better-sqlite3 was only present as context-mode's transitive build
-	// build — and nothing else. pi-hermes-memory's own lockfile carries a
+	// dependency; with context-mode retired the root lockfile must not carry a
-	// nested major that ships N-API prebuilds and needs no approval entry.
+	// root-level better-sqlite3 and allowScripts must not approve one.
+	// pi-hermes-memory's own lockfile carries a nested major that ships N-API
+	// prebuilds and needs no approval entry.
 	const rootLock = JSON.parse(await readFile(join(repositoryRoot, "package-lock.json"), "utf8")) as {
 		packages: Record<string, { version?: string }>;
 	};
 	const hermesLock = JSON.parse(await readFile(join(repositoryRoot, "pi-hermes-memory", "package-lock.json"), "utf8")) as {
 		packages: Record<string, { version?: string }>;
 	};
-	const rootSqliteVersion = rootLock.packages["node_modules/better-sqlite3"]?.version;
 	const hermesSqliteVersion = hermesLock.packages["node_modules/better-sqlite3"]?.version;
-	assert.match(rootSqliteVersion ?? "", /^\d+\.\d+\.\d+$/, "the root lockfile must pin a root-level better-sqlite3");
+	assert.equal(
+		rootLock.packages["node_modules/better-sqlite3"],
+		undefined,
+		"the root lockfile must not pin a root-level better-sqlite3 now that context-mode is gone",
+	);
 	assert.match(hermesSqliteVersion ?? "", /^\d+\.\d+\.\d+$/, "the Hermes lockfile must pin its own better-sqlite3");
 	assert.deepEqual(
-		Object.keys(packageJson.allowScripts).filter((key) => key.startsWith("better-sqlite3@")),
+		Object.keys(packageJson.allowScripts ?? {}).filter((key) => key.startsWith("better-sqlite3@")),
-		[`better-sqlite3@${rootSqliteVersion}`],
+		[],
-		"allowScripts must approve exactly the root lockfile's better-sqlite3 build",
+		"allowScripts must not approve a better-sqlite3 build without a root-level package",
-	);
-	assert.equal(
-		packageJson.allowScripts[`better-sqlite3@${hermesSqliteVersion}`],
-		undefined,
-		`Hermes uses better-sqlite3 ${hermesSqliteVersion} N-API prebuilds, so no extra install-script approval is required`,
 	);
 	assert.equal(packageJson.dependencies["@sinclair/typebox"], undefined);
 	const home = await mkdtemp(join(tmpdir(), "my-pi-extension-load-"));

tests/package-files.test.ts

@@ -0,0 +1,94 @@
+import assert from "node:assert/strict";
+import { execFileSync } from "node:child_process";
+import { readFileSync } from "node:fs";
+import { dirname, join, posix, resolve } from "node:path";
+import { fileURLToPath } from "node:url";
+import test from "node:test";
+
+const repositoryRoot = resolve(dirname(fileURLToPath(import.meta.url)), "..");
+
+const packageJson = JSON.parse(readFileSync(join(repositoryRoot, "package.json"), "utf8")) as {
+	dependencies: Record<string, string>;
+	pi: { extensions: string[] };
+};
+
+const packOutput = execFileSync("npm", ["pack", "--dry-run", "--json", "--ignore-scripts"], {
+	cwd: repositoryRoot,
+	encoding: "utf8",
+});
+const packResult = JSON.parse(packOutput.slice(packOutput.indexOf("["))) as Array<{
+	files: Array<{ path: string }>;
+}>;
+const packedFiles = new Set(packResult[0].files.map((entry) => entry.path));
+
+test("the packed bundle ships no nested node_modules or build output", () => {
+	assert.ok(packedFiles.size > 0, "npm pack --dry-run must report a non-empty file list");
+	for (const filePath of packedFiles) {
+		const segments = filePath.split("/");
+		assert.equal(segments.includes("node_modules"), false, `${filePath} must not ship a nested node_modules tree`);
+		assert.equal(segments.includes("dist"), false, `${filePath} must not ship build output`);
+	}
+});
+
+test("every declared extension entry and file: dependency manifest is inside the packed bundle", () => {
+	for (const entry of packageJson.pi.extensions) {
+		const bundled = entry.replace(/^\.\//u, "");
+		assert.ok(packedFiles.has(bundled), `extension entry ${entry} must be inside the packed bundle`);
+	}
+	for (const [name, specifier] of Object.entries(packageJson.dependencies)) {
+		const match = /^file:\.\/(.+)$/u.exec(specifier);
+		if (!match) continue;
+		const manifest = `${match[1]}/package.json`;
+		assert.ok(packedFiles.has(manifest), `${name} is a file: dependency — ${manifest} must be packed`);
+	}
+});
+
+test("relative import specifiers inside packed sources resolve to packed files", () => {
+	const specifierPatterns = [
+		/\bfrom\s+["'](\.{1,2}\/[^"']+)["']/gu,
+		/(?<![\w$.])import\s*\(\s*["'](\.{1,2}\/[^"']+)["']\s*\)/gu,
+		/\bnew\s+URL\s*\(\s*["'](\.{1,2}\/[^"']+)["']\s*,\s*import\.meta\.url\s*\)/gu,
+	];
+	const isPacked = (resolved: string): boolean => {
+		if (packedFiles.has(resolved)) return true;
+		const candidates = [`${resolved}.ts`, `${resolved}.js`, `${resolved}.mjs`, `${resolved}/index.ts`];
+		if (candidates.some((candidate) => packedFiles.has(candidate))) return true;
+		// TypeScript sources import sibling ".js"/".mjs" specifiers that map to ".ts" files.
+		const swapped = resolved.replace(/\.(js|mjs|cjs)$/u, ".ts");
+		if (swapped !== resolved && packedFiles.has(swapped)) return true;
+		// new URL("../config/prompt/", import.meta.url) style directory references.
+		for (const filePath of packedFiles) {
+			if (filePath.startsWith(`${resolved}/`)) return true;
+		}
+		return false;
+	};
+	const unresolved: string[] = [];
+	for (const filePath of packedFiles) {
+		if (!/\.(ts|js|mjs)$/u.test(filePath)) continue;
+		const segments = filePath.split("/");
+		if (/\.test\.(ts|js|mjs)$/u.test(filePath) || segments.includes("test") || segments.includes("tests")) continue;
+		const source = readFileSync(join(repositoryRoot, filePath), "utf8");
+		for (const pattern of specifierPatterns) {
+			pattern.lastIndex = 0;
+			for (const match of source.matchAll(pattern)) {
+				const resolved = posix.normalize(posix.join(posix.dirname(filePath), match[1])).replace(/\/$/u, "");
+				if (resolved === ".." || resolved.startsWith("../")) continue;
+				if (resolved.split("/").includes("node_modules")) continue;
+				if (!isPacked(resolved)) unresolved.push(`${filePath} -> ${match[1]}`);
+			}
+		}
+	}
+	assert.deepEqual(unresolved, [], "packed sources reference files missing from the bundle");
+});
+
+test("prompt config and lifecycle scripts are inside the packed bundle", () => {
+	const tracked = execFileSync("git", ["ls-files", "config/prompt"], {
+		cwd: repositoryRoot,
+		encoding: "utf8",
+	})
+		.split("\n")
+		.filter(Boolean);
+	for (const filePath of [...tracked, "install.sh", "update.sh", "uninstall.sh", "configure.sh", "scripts/lib.sh"]) {
+		assert.ok(packedFiles.has(filePath), `${filePath} must be inside the packed bundle`);
+	}
+});

tests/system-prompt.test.ts

@@ -29,10 +29,6 @@ import systemPromptExtension, {
 	getSystemPromptDumpPath,
 } from "../extensions/system-prompt.ts";
 import { guardExtensionApi } from "../extensions/prompt-guard.ts";
-import {
-	CONTEXT_MODE_ANCHOR_PREFIX,
-	filterContextModeMessages,
-} from "../extensions/context-mode-filter.ts";
 import { BUNDLE_GROUP_DEFINITIONS } from "../pi-tool-search/extensions/bundle-groups.ts";
 import { BUNDLE_TOOL_SEARCH_DEFAULTS } from "../pi-tool-search/extensions/config.ts";
 
@@ -164,7 +160,6 @@ test("prompt config carries no retired guidance phrasing", () => {
 		"read/grep",
 		"replace_text",
 		"Read/edit files",
-		"ctx_fetch_and_index then ctx_search",
 	];
 	for (const file of ["preamble.md", "tools.jsonc", "routing.jsonc"]) {
 		const source = readFileSync(join(promptConfigDir, file), "utf8");
@@ -267,8 +262,8 @@ test("bundled routing rules select the same guidance the retired tool-routing ex
 	assert.match(ssh, /only when the user explicitly names an imported host/);
 	assert.match(ssh, /command-local directory change/);
 
-	const chrome = renderRouting(routing.rules, ["chrome_snapshot", "ctx_execute_file"]);
+	const chrome = renderRouting(routing.rules, ["chrome_snapshot", "overflow_query"]);
-	assert.match(chrome, /immediately analyze the returned workspace-relative capture path/);
+	assert.match(chrome, /return the page snapshot JSON directly/);
 
 	const minimal = renderRouting(routing.rules, ["read"]);
 	assert.doesNotMatch(minimal, /codegraph_explore first/);
@@ -646,7 +641,7 @@ test("prompt guard runs context results through the provided contextFilter", asy
 test("prompt guard returns the handler result untouched when the filter changes nothing", async () => {
 	const { pi, handlers } = makeRecordingPi();
 	const guarded = guardExtensionApi(pi as unknown as ExtensionAPI, {
-		contextFilter: filterContextModeMessages,
+		contextFilter: (_before, after) => after,
 	});
 	const expected = { messages: [{ role: "user", content: "plain" }] };
 
@@ -662,67 +657,6 @@ test("prompt guard passes unrelated API members through to the real pi object",
 	assert.equal((guarded as unknown as { someOtherApi: () => string }).someOtherApi(), "passthrough");
 });
 
-// ── context-mode filter ─────────────────────────────────────────────────────
-
-test("context-mode filter drops only the routing-anchor part", () => {
-	const before = [{ role: "user", content: "real" }];
-	const anchor = `${CONTEXT_MODE_ANCHOR_PREFIX} ctx_batch_execute > ctx_execute`;
-	const after = [
-		...before,
-		{ role: "user", content: `${anchor}\n\n<active_memory>\nx\n</active_memory>` },
-	];
-	const filtered = filterContextModeMessages(before, after);
-	assert.equal(filtered.length, 2);
-	assert.equal(filtered[1].content, "<active_memory>\nx\n</active_memory>");
-});
-
-test("context-mode filter removes the appended message when only the anchor remains", () => {
-	const before = [{ role: "user", content: "real" }];
-	const after = [
-		...before,
-		{ role: "user", content: `${CONTEXT_MODE_ANCHOR_PREFIX} only anchor` },
-	];
-	assert.deepEqual(filterContextModeMessages(before, after), before);
-});
-
-test("context-mode filter returns the same reference when nothing changed", () => {
-	const before = [{ role: "user", content: "real" }];
-	const after = [...before, { role: "user", content: "no anchor here" }];
-	assert.equal(filterContextModeMessages(before, after), after);
-	assert.equal(filterContextModeMessages(after, after), after);
-});
-
-test("context-mode filter ignores appended non-user or non-string messages", () => {
-	const before = [{ role: "user", content: "real" }];
-	const after = [
-		...before,
-		{ role: "assistant", content: `${CONTEXT_MODE_ANCHOR_PREFIX} not ours` },
-		{ role: "user", content: [{ type: "text", text: "structured" }] },
-	];
-	assert.equal(filterContextModeMessages(before, after), after);
-});
-
-// ── Pinned upstream text ────────────────────────────────────────────────────
-
-test("context-mode upstream still appends the single anchor-bearing user message", async () => {
-	const packageJson = JSON.parse(
-		await readFile(join(repositoryRoot, "package.json"), "utf8"),
-	) as { dependencies: Record<string, string> };
-	assert.match(
-		packageJson.dependencies["context-mode"] ?? "",
-		/^\d+\.\d+\.\d+$/,
-		"the context-filter pin requires an exact upstream version",
-	);
-	const upstream = await readFile(
-		join(repositoryRoot, "node_modules/context-mode/build/adapters/pi/extension.js"),
-		"utf8",
-	);
-	assert.ok(upstream.includes('parts.push("context-mode active. Hierarchy:'));
-	assert.ok(upstream.includes('_pendingContext = extraParts.join("\\n\\n")'));
-	assert.ok(upstream.includes("event.messages.push({"));
-	assert.ok(upstream.includes('role: "user"'));
-});
-
 // ── /dump-system-prompt ─────────────────────────────────────────────────────
 
 test("dump-system-prompt writes the last composed prompt under .pi-debug", async () => {

tests/web-search.test.ts

@@ -5,9 +5,12 @@ import test from "node:test";
 const packageJson = JSON.parse(await readFile(new URL("../package.json", import.meta.url), "utf8")) as {
 	dependencies: Record<string, string>;
 	files: string[];
-	pi: { extensions: string[]; skills: string[] };
+	pi: { extensions: string[]; skills?: string[] };
 };
 
+const filesIncludePackageDir = (files: string[], packageDir: string): boolean =>
+	files.some((entry) => entry === packageDir || entry.startsWith(`${packageDir}/`));
+
 test("pi-web-search is the single search surface and reads credentials from auth.json", () => {
 	const entries = packageJson.pi.extensions;
 	assert.ok(entries.includes("./pi-web-search/index.ts"), "the bundle must load pi-web-search");
@@ -27,7 +30,7 @@ test("pi-web-search is the single search surface and reads credentials from auth
 
 test("pi-web-search ships as a maintained file: dependency without upstream search packages", () => {
 	assert.equal(packageJson.dependencies["pi-web-search"], "file:./pi-web-search");
-	assert.ok(packageJson.files.includes("pi-web-search"));
+	assert.ok(filesIncludePackageDir(packageJson.files, "pi-web-search"));
 	assert.equal(packageJson.dependencies["@tavily/pi-extension"], undefined);
 	assert.equal(packageJson.dependencies["@keenable/pi-search"], undefined);
 	assert.equal(packageJson.dependencies["pi-mcp-adapter"], undefined, "Exa is a direct client now — no MCP adapter");
@@ -42,7 +45,7 @@ test("no extension in the bundle loads the MCP adapter or upstream search extens
 });
 
 test("the keenable skill path is gone with the upstream package", () => {
-	for (const skillPath of packageJson.pi.skills) {
+	for (const skillPath of packageJson.pi.skills ?? []) {
 		assert.equal(skillPath.includes("@keenable"), false);
 	}
 });