메뉴
Analyze recent /define sessions to extract user preference patterns and write them to CLAUDE.md. Use when you want to learn from past define sessions, extract define patterns, improve future defines, or capture define preferences.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/doodledood/claude-code-plugins --skill learn-define-patterns명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
End-to-end autonomous execution: figure-out → define → do, chained without manual approval gates. Use when you want to define and execute without intervention during planning, when the user asks for autonomous or end-to-end work, says just build it, or asks to tend or babysit a PR.
Author-side PR lifecycle babysitter and companion to review-pr. Use when the user wants to tend an existing GitHub PR through CI, review threads, description sync, mergeability, auto-fixes, or asks to babysit a PR with or without an existing manifest.
Manifest executor. Works through Deliverables verifying every Acceptance Criterion and Global Invariant. Use when executing a manifest, running a plan, implementing a defined task, or when the user asks to run, execute, implement, or ship a manifest-backed plan.
SKILL.md 표시 중
SOC 직업 분류 기준
| name | learn-define-patterns |
| description | Analyze recent /define sessions to extract user preference patterns and write them to CLAUDE.md. Use when you want to learn from past define sessions, extract define patterns, improve future defines, or capture define preferences. |
| user-invocable | true |
User request: $ARGUMENTS
Analyze recent /define session transcripts, extract patterns in how the user approaches /define interviews (probing preferences, trade-off defaults, recurring invariants, process guidance, quality gate adjustments), and write generalizable patterns to CLAUDE.md as ## /define Preferences. Future /define sessions see these preferences automatically because CLAUDE.md is loaded into context.
Every /define session, users make the same corrections, add the same invariants, resolve the same trade-offs. Without learning, each session starts from zero. This skill closes the feedback loop: patterns from past sessions become probing hints for future ones.
| Constraint | Rule |
|---|---|
| User approval required | NEVER write to CLAUDE.md without presenting patterns to the user and getting explicit approval. |
| Merge, never overwrite | If a ## /define Preferences section already exists, merge new patterns with existing ones. Never blindly overwrite. |
| Semantic deduplication | When merging, identify patterns that say the same thing in different words and consolidate them. Don't just check for exact text matches. |
| Standard markdown only | Output uses ## headers, ### subheaders, - bullets, and <!-- date --> HTML comments. No custom syntax, no YAML, no special parsing. |
| Ask write target | Ask the user which CLAUDE.md to write to: project CLAUDE.md, user ~/.claude/CLAUDE.md, or both. Never assume. |
| Diff preview before write | Show the user exactly what will be added or changed in CLAUDE.md before writing. |
| Clean up temp files | Delete per-session analysis files from /tmp/ after aggregation is complete. |
Session JSONL files live at ~/.claude/projects/{project-path-encoded}/{session-id}.jsonl. Find recent sessions containing /define activity. If $ARGUMENTS specifies a session count, use that; otherwise use enough recent sessions for meaningful pattern signal.
No sessions found: Tell the user: "No /define sessions found in recent session history. Run a few /define sessions first, then try again."
Malformed files: Skip with a warning noting which files were skipped and why.
Use the define-session-analyzer agent to analyze each session independently. Each agent receives a session file path and an output path (/tmp/define-learn-{session-id}.md). Sessions with zero extractable patterns are normal — count them in the final summary.
The final output is a unified set of user preferences derived from all analyzed sessions.
Quality criteria for the aggregated output:
What the user sees before approving:
~/.claude/CLAUDE.md, or both## /define Preferences
### Probing Hints
- Pattern statement here <!-- 2026-03-01 -->
### Trade-off Defaults
- Pattern statement here <!-- 2026-03-01 -->
### Recurring Invariants
- Pattern statement here <!-- 2026-03-01 -->
### Process Guidance
- Pattern statement here <!-- 2026-03-01 -->
### Quality Gate Adjustments
- Pattern statement here <!-- 2026-03-01 -->
### Other
- Pattern statement here <!-- 2026-03-01 -->
When merging with an existing ## /define Preferences section: preserve all existing patterns and their date comments, deduplicate new patterns against existing ones by meaning (not just text match), add new patterns under the appropriate subcategory headers, and omit empty subcategories.
Precedence: When future /define sessions encounter a conflict between built-in task file guidance and patterns in ## /define Preferences, the user's patterns represent intentional preferences and take precedence.
Traceability: Each pattern includes an inline <!-- YYYY-MM-DD --> date comment. Users remove patterns by editing CLAUDE.md directly — no special tooling needed.
After writing, output a summary: sessions analyzed, sessions with patterns (and sessions with zero patterns), patterns extracted, patterns approved, contradictions found, and which CLAUDE.md was written to.