feat(pi-permission-system): AutoReview 改为跟随主模型的对话式工具调用审查

AutoReview:

- 审查器回到 Codex Guardian 风格的对话式审查,默认跟随会话当前主模型
  (每次审查实时读取 ctx.model);provider/model 必须同时设置才固定审查
  模型,只设其一视为配置错误并转人工。
- 结论只通过 approve / reject 两个工具给出,参数仅 reason;不强制
  toolChoice,正文文本(即使是合法 JSON)一律忽略。
- approve 自动放行;reject 转人工并在弹窗显示理由(标注模型生成)。
  超时、Provider 错误、截断、未调用工具、多次调用、参数无效等异常同样
  转人工,并附固定的系统说明;用户按 Esc 中止时直接拒绝。
- 模型调用改为 modelRegistry.streamSimple(...).result():Pi 1.3.0 面向
  扩展的 ModelRegistry 没有 completeSimple,原写法在真实环境中每次都会
  转人工。
- 对话记录保留可信度标注;工具结果在有更短的 pi-compaction 摘要时替换
  为摘要;token 估算改为非 ASCII 字符按 1 token 计;预算可配置。
- 同一条用户消息内完全相同的请求只缓存 approve 结论。
- 默认 reasoning low、总超时 30 秒、最多 2 次尝试、maxOutputTokens 4000。

删除:

- 未发布过的 TypeSafe Jev 分类器实现、DenialCircuitBreaker。
- 整个 annotator 机制(model-explanation、注册表、registerAnnotator、
  provider-timeout);annotators 配置键保留为已弃用并忽略,避免旧配置
  被拒绝。

测试:

- 新增 reviewer / transcript / 弹窗说明 / 默认配置测试;清理重复、恒真
  或只测实现细节的旧测试。
- 使用 deepseek-v4.1-flash 实测 5 个场景各 5 次,25 次结论全部符合预期。

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

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
This commit is contained in:
云服务部-叶林立
2026-10-19 01:54:17 -07:00
co-authored by Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
parent dec48dab6d
commit 7e6a8faae1
87 changed files with 2831 additions and 2658 deletions
+12 -2
View File
@@ -5,8 +5,18 @@
"doublePressToConfirm": true,
// AutoReview is the default authority for deterministic review decisions.
// Automatic review stays inactive until you name both a provider and a model
// here; until then every ask goes to the human permission prompt.
//
// The reviewer is a conversational model asked to judge one planned action
// under a bundled security policy. With no provider/model pair it follows
// the session's current model, so review quality tracks whatever you run.
// It answers by calling `approve` or `reject`; a reject — and every failure
// (timeout, provider error, malformed verdict) — defers to the human
// permission prompt with the model's reason shown. Approvals are cached
// within one user message, so the same ask does not pay for a second review.
//
// To pin a dedicated review model instead, set BOTH "provider" and "model";
// exactly one set is a misconfiguration and also defers to the human.
// "reasoning" defaults to "low" and only applies to reasoning-capable models.
"authorizerChain": ["auto-review"],
"autoReview": {
"enabled": true