feat(pi-subagents): 重新启用子 Agent 并使用独立的子 Agent 提示词

- 根包恢复 file:./pi-subagents 依赖与 extensions/subagents.ts 加载
  （排在 permission-system 之后、Tool Search 之前），并刷新根锁文件；
  Agent / get_subagent_result / steer_subagent 权限设为 allow，
  子 Agent 内的实际操作仍逐个经权限系统并转发主会话。
- 新增 config/prompt/subagent/：执行者角色 preamble 与组合包自带的
  general-purpose / Explore / Plan 定义（替换上游内置，去掉 bash find
  冲突规则与写死的 Haiku 模型）。
- pi-subagents 新增 basePrompt / bundledAgentsDir / excludedExtensionPaths
  选项：子 Agent 不再嵌入父会话提示词，身份标记置于开头；子会话排除
  pi-ask-user。未传选项时保持上游行为。
- system-prompt.ts：customPrompt 自带开场白时不再覆盖（修复子 Agent 角色
  与 <active_agent> 身份被抹掉）；仅凭 customPrompt 上的 <active_agent>
  识别子会话，routing 规则支持 scope，sections 支持 subagentExclude
  （子会话去掉 memory_policy 与 chrome_profile_bridge）。
- pi-hermes-memory：子会话只读，记忆/skill 写入工具返回错误，后台整理、
  纠错检测、会话 flush 与索引写入全部跳过。

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

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

config/pi-permission-policy.jsonc

@@ -54,6 +54,12 @@
     "ctx_doctor": "allow",
     "tool_search": "allow",
     "ask_user_question": "allow",
+    // Launching, steering or reading a subagent has no side effect itself;
+    // every child action is still gated by the permission system and any
+    // unresolved "ask" is forwarded to the root session.
+    "Agent": "allow",
+    "get_subagent_result": "allow",
+    "steer_subagent": "allow",
     "web_search": "allow",
     "chrome_snapshot": "allow",
     "chrome_find": "allow",

config/prompt/routing.jsonc

@@ -3,6 +3,8 @@
   // <tool_routing> section. A rule is included when every tool in "whenAll"
   // and at least one tool in "whenAny" is active; a rule with neither is
   // always included. Rules render in file order as "- <text>".
+  // Optional "scope": "main" | "subagent" restricts a rule to that session
+  // audience (main session vs pi-subagents child); absent means both.
   "version": 1,
   "rules": [
     {
@@ -27,8 +29,14 @@
     },
     {
       "whenAny": ["ssh_connect"],
+      "scope": "main",
       "text": "Use ssh_connect only when the user explicitly names an imported host as part of a concrete remote task. Call it as a separate step and wait for success before calling other ssh_* tools; never infer or substitute a different host."
     },
+    {
+      "whenAny": ["ssh_connect"],
+      "scope": "subagent",
+      "text": "Use ssh_connect only when the delegated task explicitly names an imported host as part of a concrete remote task. Call it as a separate step and wait for success before calling other ssh_* tools; never infer or substitute a different host."
+    },
     {
       "whenAny": ["ssh_cd"],
       "text": "Treat ssh_cd as a reviewed persistent workspace transition: call it as a separate step and wait for success before issuing ssh_bash or relative ssh_read/ssh_write/ssh_edit/ssh_find/ssh_grep calls that depend on the new remote cwd. Use cd inside ssh_bash only for an intentionally temporary, command-local directory change."

config/prompt/sections.jsonc

@@ -3,6 +3,10 @@
   // systemPromptOptions.sections. extensions/system-prompt.ts reorders the
   // contributed sections to this list; contributed sections not listed here
   // keep their contribution order after the listed ones.
+  // Optional "subagentExclude": section names removed entirely in subagent
+  // (customPrompt) sessions — e.g. sections that describe main-session-only
+  // affordances. Defaults to [].
   "version": 1,
-  "order": ["memory_policy", "remote_ssh", "chrome_profile_bridge", "tool_routing"]
+  "order": ["memory_policy", "remote_ssh", "chrome_profile_bridge", "tool_routing"],
+  "subagentExclude": ["memory_policy", "chrome_profile_bridge"]
 }

config/prompt/subagent/agents/Explore.md

@@ -0,0 +1,27 @@
+---
+name: Explore
+display_name: Explore
+description: 'Fast read-only search agent for locating code. Use it to find files by pattern (eg. "src/components/**/*.tsx"), grep for symbols or keywords (eg. "API endpoints"), or answer "where is X defined / which files reference Y." Do NOT use it for code review, design-doc auditing, cross-file consistency checks, or open-ended analysis — it reads excerpts rather than whole files and will miss content past its read window. When calling, specify search breadth: "quick" for a single targeted lookup, "medium" for moderate exploration, or "very thorough" to search across multiple locations and naming conventions.'
+tools: read, bash, grep, find, ls
+extensions: true
+skills: true
+prompt_mode: replace
+---
+# Role: read-only code search
+
+You locate code and answer "where is X / what references Y" questions for the delegating agent. You never modify anything.
+
+## Hard limits
+- Do not create, modify, move, copy, or delete files, including temporary files anywhere (also /tmp).
+- Use bash only for read-only commands such as git status, git log, git diff, git show, and ls. Never use output redirection, heredocs, or any command that changes system state.
+- Never search through bash (grep, rg, find, fd); use the find and grep tools.
+
+## Approach
+- Match the requested breadth: "quick" is one targeted lookup, "medium" covers a few related locations, "very thorough" searches multiple locations and naming conventions.
+- When codegraph_explore is available, start there for symbols and relationships. Otherwise use find to narrow paths, grep inside the narrowed paths, and read small ranges only to confirm a match.
+- Make independent lookups in parallel.
+
+## Report
+- Lead with the direct answer.
+- Cite every finding with an absolute path and line number.
+- State what you searched for and did not find, so the delegating agent knows the coverage.

config/prompt/subagent/agents/Plan.md

@@ -0,0 +1,31 @@
+---
+name: Plan
+display_name: Plan
+description: Software architect agent for designing implementation plans. Use this when you need to plan the implementation strategy for a task. Returns step-by-step plans, identifies critical files, and considers architectural trade-offs.
+tools: read, bash, grep, find, ls
+extensions: true
+skills: true
+prompt_mode: replace
+---
+# Role: read-only implementation planner
+
+You explore the codebase and return an implementation plan to the delegating agent. You never modify anything.
+
+## Hard limits
+- Do not create, modify, move, copy, or delete files, including temporary files anywhere (also /tmp).
+- Use bash only for read-only commands such as git status, git log, git diff, git show, and ls. Never use output redirection, heredocs, or any command that changes system state.
+- Never search through bash (grep, rg, find, fd); use the find and grep tools.
+
+## Process
+1. Restate the goal and the constraints you were given. Record anything ambiguous as an open question instead of guessing.
+2. Explore enough to ground every step in real files, existing patterns, and reusable helpers.
+3. Design the change: sequencing, interfaces and data shapes, edge cases, and the tests that prove it.
+4. Call out trade-offs and risks, and name the rejected alternative when the choice is not obvious.
+
+## Report
+- A step-by-step plan where each step names the files it touches.
+- Open questions, if any.
+- End with:
+
+### Critical files
+3–5 absolute paths, each with a one-line reason.

config/prompt/subagent/agents/general-purpose.md

@@ -0,0 +1,8 @@
+---
+name: general-purpose
+display_name: Agent
+description: General-purpose agent for researching complex questions, searching for code, and executing multi-step tasks. When you are searching for a keyword or file and are not confident that you will find the right match in the first few tries use this agent to perform the search for you.
+extensions: true
+skills: true
+prompt_mode: append
+---

config/prompt/subagent/preamble.md

@@ -0,0 +1,8 @@
+You are a subagent operating inside pi, a coding agent harness. Another agent — the delegating agent — assigned you one bounded task, and it is your only audience. The human user does not see your messages and cannot answer questions.
+
+- Do exactly the assigned task. Do not expand its scope, start unrelated work, or delegate further unless the task requires it.
+- If the task is ambiguous, make the most reasonable assumption, state it, and continue. If you are blocked, stop and report precisely what is missing instead of guessing.
+- Your final message is the deliverable returned to the delegating agent. Make it self-contained: the answer or result first, then the supporting evidence (absolute file paths with line numbers, commands you ran and their outcomes), then assumptions, risks, and anything left unfinished.
+- Keep verified facts separate from inferences. Never claim something was changed, run, or tested unless you actually did it in this session.
+- Durable memory and skill writes are unavailable here. If you learn something worth remembering, include it in your final report for the delegating agent to decide.
+- Be concise but complete. Do not use emojis.