{"skill":{"slug":"retrospective-agent","displayName":"Retrospective Agent","summary":"Structured retrospectives and execution-memory hygiene for OpenClaw agents. Use when the user wants a retrospective, lessons learned, self-improvement system...","description":"---\nname: retrospective-agent\ndescription: Structured retrospectives and execution-memory hygiene for OpenClaw agents. Use when the user wants a retrospective, lessons learned, self-improvement system, correction logging, weekly review, or a clean way to capture reusable execution lessons without creating hidden memory or autonomous behavior.\n---\n\n# Retrospective Agent\n\nUse this skill to capture execution lessons in a controlled, auditable way.\n\nThis skill exists to improve how the agent works over time.\nIt does not create a second factual memory system, rewrite identity, or invent autonomy.\n\n## Core principles\n\n- Keep factual continuity in existing memory files\n- Keep execution lessons separate and scoped\n- Prefer reports and recommendations over automatic changes\n- Promote patterns only after repeated evidence\n- Never infer preferences from silence\n- Never rewrite persona, config, or outbound behavior on your own\n\n## Memory split\n\n### Use existing memory for\n- facts\n- events\n- decisions\n- dates\n- people\n- open tasks\n\nExamples:\n- `memory/YYYY-MM-DD.md`\n- agent `MEMORY.md`\n- project `README.md`\n\n### Use retrospective-agent files for\n- repeated corrections\n- workflow improvements\n- tool failure patterns\n- success patterns worth repeating\n- project or domain execution lessons\n\n## Storage\n\nSkill files live in:\n- `workspace/skills/retrospective-agent/`\n\nOperational data lives in:\n- `workspace/ops/retrospective-agent/`\n\nExpected first-pass files:\n- `workspace/ops/retrospective-agent/corrections.md`\n- `workspace/ops/retrospective-agent/weekly/`\n- `workspace/ops/retrospective-agent/domains/`\n- `workspace/ops/retrospective-agent/projects/`\n- `workspace/ops/retrospective-agent/templates/`\n\nIf the ops folder or expected files do not exist, create only the minimum needed for the current task.\nDo not create extra files \"just in case\".\n\n## Triggers\n\nUse this skill when:\n- the user asks for a retrospective or lessons learned\n- a multi-step task ends and a short retro would be useful\n- the user gives a reusable correction\n- a process or tool fails in a reusable way\n- a project needs scoped lessons for future work\n- a weekly review is requested\n\nDo not use this skill for:\n- one-off instructions with no reusable lesson\n- customer messaging drafts\n- sensitive personal profiling\n- fake automation or hidden monitoring claims\n\n## Operating modes\n\n### 1. Post-task retrospective\n\nUse after meaningful work.\n\nOutput:\n- what went well\n- what went wrong\n- what to repeat\n- what to change next time\n- whether anything deserves logging\n\nKeep it short and operational.\n\n### 2. Correction logging\n\nUse when an explicit correction reveals a reusable lesson.\n\nWorkflow:\n1. capture the exact correction\n2. classify it\n3. choose scope: project, domain, or global execution lesson\n4. append a concise entry if warranted\n5. recommend promotion only after repeated evidence\n\n### 3. Weekly retrospective\n\nUse on demand or when a scheduled review is explicitly requested.\n\nOutput:\n- recurring wins\n- recurring misses\n- repeated patterns\n- candidate updates to memory, README files, or skills\n\n## Scope hierarchy\n\nMost specific wins:\n1. project\n2. domain\n3. global execution lesson\n\nIf scope is unclear, prefer domain over global.\nIf still unclear, say so.\n\n## Promotion model\n\nUse conservative states:\n- observed\n- repeated\n- candidate rule\n- confirmed rule\n\nSuggested threshold:\n- 1 occurrence: observed\n- 2 occurrences: repeated\n- 3 occurrences: candidate rule\n\nDo not silently promote a candidate into durable agent behavior everywhere.\nRecommend the promotion and ask when confirmation matters.\n\n## Guardrails\n\nNever:\n- rewrite `SOUL.md`\n- rewrite `IDENTITY.md`\n- rewrite `USER.md`\n- patch config\n- send messages\n- install companion skills without approval\n- infer preferences from silence\n- store credentials, secrets, or sensitive personal data\n- claim autonomous monitoring unless a real scheduler exists\n\n## Workflow references\n\nRead these only when needed:\n- `references/workflow.md`\n- `references/promotion-rules.md`\n- `references/boundaries.md`\n\nUse templates from:\n- `assets/templates/post-task-retro.md`\n- `assets/templates/weekly-retro.md`\n- `assets/templates/lesson-entry.md`\n\n## Style\n\nBe honest, compact, and boring in a good way.\nAvoid AGI theater, inflated claims, and vague self-improvement language.\nPrefer operational wording like \"lesson\", \"pattern\", \"correction\", and \"recommended update\" over dramatic wording like \"optimize myself\" or \"evolve\".\n\n## Output rule\n\nLead with the useful retrospective or lesson.\nDo not narrate the framework unless the user asks.\n","tags":{"latest":"1.0.0"},"stats":{"comments":0,"downloads":767,"installsAllTime":3,"installsCurrent":3,"stars":1,"versions":1},"createdAt":1773513336485,"updatedAt":1778491910241},"latestVersion":{"version":"1.0.0","createdAt":1773513336485,"changelog":"Initial release","license":"MIT-0"},"metadata":null,"owner":{"handle":"sebclawops","userId":"s17ag4byms181tn4krsyt175tn83ghvh","displayName":"Sebastian The Claw","image":"https://avatars.githubusercontent.com/u/262026950?v=4"},"moderation":null}