メニュー
Use when reviewing a skill/agent/rule and you need to understand why it looks the way it does. Shows edit history, churn, stability, and fix-after-feat patterns. Keywords: history, evolution, changes, timeline, churn, lifecycle.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | evolution |
| description | Use when reviewing a skill/agent/rule and you need to understand why it looks the way it does. Shows edit history, churn, stability, and fix-after-feat patterns. Keywords: history, evolution, changes, timeline, churn, lifecycle. |
| argument-hint | <file-path> [commit1..commit2] |
| disable-model-invocation | false |
| user-invocable | true |
| allowed-tools | ["Read","Grep","Glob","Bash(git:*)"] |
You are running evolution — tracking how a definition file changed over time. Target: $ARGUMENTS
Parse args → Gather history → Analyze patterns → Report timeline + stability
Extract file path and optional commit range from $ARGUMENTS. Formats: <file-path>, <file-path> commit1..commit2, or <file-path> commit1 commit2.
Verify file exists in working tree or git history. If not found, error: "File not found in current tree or git history."
# Timeline
git log --follow --format="%H|%ai|%s" -- <file-path>
# Line deltas
git log --follow --stat --format="%H" -- <file-path>
Extract for each commit: hash, date, type (from conventional prefix), message, line delta (+/-).
Compute:
Detect fix-after-feat: consecutive feat → fix commits within 7 days (rushed changes).
Classify stability by days since last edit:
Compute edit frequency: (last date - creation date) / (edit count - 1). Report "single edit" if only one.
Trend (last 5 edits): net line change. growing (>+50), shrinking (<-50), or stable (-50 to +50).
Format output in pipe format so downstream skills (/curate, /assess, /critique) can consume the findings:
## Evolution: <file-path>
**Source**: /evolution
**Input**: <file-path> [commit-range if provided]
**Pipeline**: (none — working from direct input)
### Items (N)
1. **Stability: <stable|active|volatile>** — <N days since last edit>, <N total edits>
- source: git log -- <file-path>
- confidence: CONFIRMED
2. **Churn: <added> added, <removed> removed** — across <N> edits since <creation-date>
- source: git log --stat -- <file-path>
- confidence: CONFIRMED
3. **Fix-after-feat: <N instances | none>** — [list commit pairs if any, or "no rushed fix cycles detected"]
- source: <commit-hash-1>, <commit-hash-2>
- confidence: CONFIRMED
4. **Trend: <growing|shrinking|stable>** — net <+/-N lines> over last 5 edits
- source: git log -5 -- <file-path>
- confidence: CONFIRMED
[Additional items for notable patterns: large single-commit rewrites, long dormant periods followed by burst activity, rename history, etc. Only include if detected — do not pad with empty observations.]
### Timeline
| Date | Commit | Type | Message | Delta |
|------|--------|------|---------|-------|
[Most recent first, all commits]
### Summary
[1-2 sentences synthesizing stability, churn trend, and any concerning patterns. Examples: "Stable definition with minimal churn — 120 days since last edit, no fix-after-feat cycles." or "Actively evolving — 8 edits in 45 days with 2 fix-after-feat cycles suggesting the design is still settling."]
If commit range provided, append diff (or --stat if >500 lines).
Interactive code review as a panel discussion: an implementer agent defends decisions while a reviewer surfaces findings, with specialist reviewers added based on what the diff touches. Use when you want findings discussed in real time rather than a static report. Keywords: PR review, code review, meeting, panel review, interactive review, review discussion.
Use when starting a new project or adding Claude Code support to an existing one. Sets up CLAUDE.md, hooks, rules, and generates tailored agents. Keywords: bootstrap, scaffold, setup, new project, init, kickstart.
Use when you have a goal but don't know which skill fits. Routes to the right skill or pipeline. Keywords: do, run, auto, go, execute, start, help me.
Use when you need to research, analyze, compare, plan, or structure thinking and aren't sure which primitive to reach for. Chains operations (gather, distill, rank, filter, verify, etc.) automatically based on your goal. Prefer this over individual primitives for multi-step work.
Use when an agent's performance is uneven and you want systematic improvement. Runs diagnose -> challenge-gen -> challenge-run -> update learnings in a loop. Works with or without a team. Keywords: active-learn, training, adversarial, improve, agent, loop, capability, tuning, solo.
Use when starting a project, forming a team for long-horizon work, or upgrading from ad-hoc agent dispatch. Creates persistent roles with ownership and learnings that improve across sessions. Keywords: team, project, setup, roles, persistent, agents, staff, learning.