1273984347
- 5 skills
- 0 followers
- 17 hours ago last updated
- ▌ Deep Review Loop · 1273984347 bundleEnforces a 5-round deep review loop (R0-R3 + V1-V5) with anti-overfit protection for plans, specs, skills, and long documents. Trigger when writing or modifying these artifacts, after batch fixes (>10 items), when the user says "deep review / 复检 / 收敛 / DRL", or when false convergence is suspected — even without explicit keywords. Do not trigger for one-off single-file edits with no structural risk, plain coding/completion without written artifacts, or casual Q&A. 5 轮深度复检闭环（R0-R3 + V1-V5）+ 4 层过拟合防护。编写/修改方案、规范、skill、长文档时、批量修复 （>10 项）后、用户要求「复检/收敛/DRL」、或怀疑假收敛时触发（即使未点名）。 不触发：无结构风险的一次性单文件小改、不涉及书面产物的常规编码/补全、闲聊问答。
- ▌
- ▌ Agent Session Loop · 1273984347 bundleManages the full agent session lifecycle as a single pipeline: deep review (审查) → wrap-up (收尾) → evolution (沉淀). Use at session end, after batch fixes or large docs, when the user says "收尾 / wrap up / 复检 / 复盘 / retro", or when docs and code mismatch and the session needs a complete close-out with memory sedimentation. 将 Agent 会话完整生命周期整合为一条流水线：深度复检（审查）→ 7 步收尾 → 复盘沉淀。 会话收尾、批量修复/长文档完成后、用户说「收尾/复检/复盘/retro」、或文档与代码不一致需完整闭环时使用。 Do not trigger for one-off single-file edits or casual Q&A, or when the user asks for a standalone deep review or retro — use deep-review-loop / self-evolution instead.
- ▌ Mem Wrap Up · 1273984347 bundleEnforces a 7-step session wrap-up pipeline: memory health check, 5-phase memory audit, fileCount sync, doc sync spot-check, heuristic sedimentation, 4-step verify, and memory-layer sync + deep-review-loop. Trigger at session end, when the user says "wrap up / 收尾 / 继续", when a workflow has been sedimented, or when docs and code mismatch — even without the exact words. Do not trigger for deep review of written artifacts (use deep-review-loop) or retro/sedimentation analysis (use self-evolution). 7 步 session 收尾流水线：memory 健康检查 → 5 phase 审计 → fileCount 同步 → 文档同步 spot-check → 复利经验沉淀 → 4 步 verify → memory 层同步 + 深度复检。session 收尾、用户说「收尾/继续」、 工作流已沉淀、或文档与代码不一致时触发（即使未点名）。 不触发：书面产物深度复检（用 deep-review-loop）、复盘/沉淀分析（用 self-evolution）。
- ▌ Self Evolution · 1273984347 bundleRuns retro after task completion. Quick mode = 3-question self-check. Full mode = 11-dimension deep analysis with knowledge upgrade and action-item triage. Trigger when the user says "全面复盘 / 周汇总 / retro / 记住这个 / capture / 经验沉淀" or after task completion — even without explicit keywords. Do not trigger for deep review of written artifacts (use deep-review-loop) or session wrap-up (use mem-wrap-up). 任务完成后的复盘 skill。快速模式 = 3 问自检；全面模式 = 11 维度深度分析 + 知识层升级 + 行动项分流。 用户说「全面复盘/周汇总/retro/记住这个/capture/经验沉淀」或任务完成时触发（即使未点名）。 不触发：书面产物深度复检（用 deep-review-loop）、session 收尾（用 mem-wrap-up）。