菜单
Audits existing skills against best practices, identifies improvement opportunities, and checks if new skills are warranted by recurring patterns. Runs daily.
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。
Tags transcript message threads with semantic concepts (roast, game-recommendation, etc.) so past conversations can be searched by idea, not just keywords.
Captures notable interactions missed during live conversation — new person info, game preferences, boundaries, jokes that landed/flopped, relationship insights.
The single owner of the hypothesis lifecycle — creates, tests, promotes, and kills hypotheses. Uses extensions as programmatic data collectors and reviews their output.
Consolidates impression files by applying weight decay, removing redundant entries, and enforcing per-user limits. Maintains relationship file summaries.
Consolidates old memories by removing duplicates, promoting recurring patterns to learned-patterns.md or relationship files, and archiving outdated entries.
Identifies behavioral patterns from recent conversations and reaction feedback, updates learned-patterns.md when 3+ observations confirm a pattern.
| name | skill-maintenance |
| description | Audits existing skills against best practices, identifies improvement opportunities, and checks if new skills are warranted by recurring patterns. Runs daily. |
| allowed-tools | Read, Edit, Write, Glob, Grep, WebFetch |
Periodically audit skills for quality, compress verbose ones, and check if new skills are needed.
Cooldown: 24 hours
WebFetch https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices — scan for any updated guidance on skill structure, token budgets, progressive disclosure, or description quality.
List local/skills/ and .claude/skills/. Pick the skill you haven't reviewed longest (or one you've noticed issues with). Read its SKILL.md.
allowed-tools in frontmatter: Not a markdown section?If improvements found in local/skills/, make the edit directly. If a framework skill in .claude/skills/ needs work, note it in today's memory file for the developer.
Read agent-data/learned-patterns.md — any pattern with 3+ occurrences that involves repeated multi-step actions? If so, consider creating a skill (invoke skill-creation first for the format reference).
If nothing needs attention, stop. Don't keep reading files hoping to find something.