菜单
Capture a durable lesson every time the human corrects the agent — or a task escalates models — so the same mistake doesn't recur. Use immediately after any user correction and after any model escalation. Also use at session start to load relevant past lessons.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/agentmc15/polytropos --skill lessons-loop命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Run Copilot CLI work through the Goliath five-role pipeline, with the architect as planner and explicit model fallbacks for implementation, test authoring, verification, review, and red-team checks.
Have Fable 5 do the expensive meta-work once — deep plan a complex task (codebase review, greenfield project, large migration) AND build the execution kit (task briefs, model-pinned subagents, skills, verification loops) so Opus/Sonnet can execute at near-Fable quality. Use when the user says "plan this with Fable", "architect this", "review this codebase", or escalates a complex task.
Run an execution kit produced by /polytropos:architect — loop through TASKS.md, dispatching each task to the kit's model-pinned subagents, verifying, and updating state. Use when the user says to execute/continue/resume a kit or plan. Args: the kit slug (optional if only one kit is active).
基于 SOC 职业分类
正在显示 SKILL.md
| name | lessons-loop |
| description | Capture a durable lesson every time the human corrects the agent — or a task escalates models — so the same mistake doesn't recur. Use immediately after any user correction and after any model escalation. Also use at session start to load relevant past lessons. |
Vendored from aesop (github:agentmc15/aesop) registry/skills/lessons-loop at commit 5506617,
with a Copilot-harness routing category added. A prompted-Reflexion pattern: the model forgets
between runs, so lessons live on disk. After a correction, write down the pattern and a rule
that prevents it; review lessons at session start. Iterate until the mistake rate drops.
tasks/lessons.md an entry with: the failure pattern, the
lesson (a rule for next time), and the contexts it applies to.tasks/lessons.md; surface the lessons relevant to the current
task.A misroute is a correction too. Record an entry with "applies_to": ["routing"] whenever:
lesson-candidate (routing): lines in the
kit's NOTES.md); orroute agent can apply it at
session start. Tiers and model ids come from data/pricing.copilot.json at run time; never
bake a price or a model ranking into a lesson.{"date": "2026-07-01", "failure_pattern": "pinned a cheap-tier model for a multi-file refactor; verify failed twice and the task escalated to strong", "lesson": "multi-file refactors start at the strong tier", "applies_to": ["routing"]}