| name | skill-session-review |
| description | Analyze how an existing agent skill was used in past Claude Code, Grok, Codex, and Oh My Pi conversations. Use when the user asks to 分析某 skill 的使用情况, 这个 skill 用着有什么问题, 根据历史对话反馈改进 this skill, review skill sessions, or write a session-review report. Writes reports/skill-session-review and a copyable qiaomu-meta handoff prompt. Do not use for skill usage counts, unused-skill cleanup, resuming a session, optimizing CLAUDE.md, creating a new skill, or applying qiaomu-meta edits.
|
| version | 0.1.0 |
| category | developer-tools-integrations |
| tags | ["skills","session-review","claude-code","grok","codex","oh-my-pi","feedback"] |
| argument-hint | [skill-name-or-path] |
| allowed-tools | Read, Glob, Grep, Write, Bash(python *), Bash(py *), Bash(git rev-parse *), Bash(git check-ignore *) |
| metadata | {"owner":"lyh","review_cadence":"quarterly"} |
Skill Session Review
In the commands below, <skill-dir> is this skill's base directory, announced when the skill loads. Substitute the literal path. On Windows, py -3 may replace python.
Review how a named existing skill was followed in real sessions. Persist a feedback report. Do not edit the target skill. Do not call qiaomu-meta to apply changes.
Hard gates
- Do not write inside the target skill directory.
- Do not emit
diff.patch or edit SKILL.md of the target.
- Do not run qiaomu-meta as part of this skill.
- Write only via the helpers below: the report file, the repo-root
.gitignore exact line, and helper --input temp files.
- Do not print full private chats in the conversation. The report uses short excerpts only.
Workflow
- Resolve the target. Path → that instance. Name → if more than one
SKILL.md, list paths and stop.
- Resolve the current repo root:
git rev-parse --show-toplevel or an explicit root the user gave.
- Scan sessions:
python "<skill-dir>/scripts/scan_invocations.py" --skill-name <name> [--skill-path <abs>] --scope global --repo-root <abs>
Use --scope cwd only when the user asked to limit to this repository. Read invocation signals. Treat loaded and available as coverage, not as required-change evidence.
- Read the target
SKILL.md (read-only) and invoked session slices. Fill findings per finding contract. Promote a pattern only when at least two invoked sessions show it.
- Write the report:
python "<skill-dir>/scripts/write_session_review.py" --repo-root <abs> --skill-name <name> [--skill-path <abs>] --input <utf8-md>
Windows: prefer --input over stdin. Template: report template. The helper appends reports/skill-session-review/ to the repo-root .gitignore when that exact line is missing. Do not git add.
- Chat output: one conclusion line, the report path, and one fence with a filled . Do not paste the SSR table. Do not write 见上一条消息.
