Skip to main content
autonomous-loops Patterns and architectures for autonomous Claude Code loops â from simple sequential pipelines to RFC-driven multi-agent DAG systems.
Ir a la instalación Skills Marketplace Descubre y explora habilidades de IA creadas por la comunidad.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Mostrar detalles del prompt Un comando directo omite el prompt de revisión. Revisa el origen antes de ejecutarlo.
npx skills add https://github.com/linnefromice/claude-code-workspace --skill autonomous-loopsEl comando permanece en una sola lÃnea. Desplázate horizontalmente para revisarlo antes de copiarlo.
¿Prefieres una copia local? Descarga los archivos que SkillsMP tiene disponibles ahora.
Descargar Zip Descargando... Ocupaciones relacionadas SOC
Basado en la clasificación ocupacional SOC
name autonomous-loops description Patterns and architectures for autonomous Claude Code loops â from simple sequential pipelines to RFC-driven multi-agent DAG systems. origin ECC
Autonomous Loops Skill
äºææ§ã«é¢ãã泚æïŒv1.8.0ïŒ: autonomous-loops ã¯1ãªãªãŒã¹åä¿æãããŸãã
æ£èŠã®ã¹ãã«å㯠continuous-agent-loop ã«ãªããŸãããæ°ããã«ãŒãã¬ã€ãã³ã¹ã¯
ãã¡ãã§äœæããŠãã ãããæ¢åã®ã¯ãŒã¯ãããŒãå£ããªãããããã®ã¹ãã«ã¯åŒãç¶ã
å©çšå¯èœã§ãã
Claude Code ãèªåŸçã«ã«ãŒãã§å®è¡ããããã®ãã¿ãŒã³ãã¢ãŒããã¯ãã£ããªãã¡ã¬ã³ã¹å®è£
ã§ããã·ã³ãã«ãª claude -p ãã€ãã©ã€ã³ãããRFC é§åã®ãã«ããšãŒãžã§ã³ã DAG ãªãŒã±ã¹ãã¬ãŒã·ã§ã³ãŸã§ãã«ããŒããŸãã
䜿çšã¿ã€ãã³ã°
人éã®ä»å
¥ãªãã«åäœããèªåŸçãªéçºã¯ãŒã¯ãããŒãã»ããã¢ããããå Žå
åé¡ã«é©ããã«ãŒãã¢ãŒããã¯ãã£ã®éžæïŒã·ã³ãã« vs è€éïŒ
CI/CD ã¹ã¿ã€ã«ã®ç¶ç¶çéçºãã€ãã©ã€ã³ã®æ§ç¯
ããŒãžèª¿æŽä»ãã®äžŠåãšãŒãžã§ã³ãã®å®è¡
ã«ãŒãã®ã€ãã¬ãŒã·ã§ã³éã§ã®ã³ã³ããã¹ãæ°žç¶åã®å®è£
èªåŸã¯ãŒã¯ãããŒãžã®å質ã²ãŒããšã¯ãªãŒã³ã¢ãããã¹ã®è¿œå
Loop Pattern Spectrum
æãã·ã³ãã«ãªãã®ããæãé«åºŠãªãã®ãŸã§:
Implementer ã¹ãããåŸã®å質ã¯ãªãŒã³ã¢ãã
1. Sequential Pipeline (claude -p) æãã·ã³ãã«ãªã«ãŒãã§ãã æ¥åžžã®éçºããã³ã€ã³ã¿ã©ã¯ãã£ã㪠claude -p ã³ãŒã«ã®ã·ãŒã±ã³ã¹ã«å解ããŸããåã³ãŒã«ã¯æ確ãªããã³ãããæã€éäžããã¹ãããã§ãã
Core Insight
ãã®ãããªã«ãŒããæ§ç¯ã§ããªããªããã€ã³ã¿ã©ã¯ãã£ãã¢ãŒãã§ã LLM ã«ã³ãŒããä¿®æ£ãããããšããã§ããªããšããããšã§ãã
claude -p ãã©ã°ã¯ããã³ããä»ãã§ãã³ã€ã³ã¿ã©ã¯ãã£ãã« Claude Code ãå®è¡ããå®äºæã«çµäºããŸããã³ãŒã«ããã§ã€ã³ããŠãã€ãã©ã€ã³ãæ§ç¯ããŸã:
#!/bin/bash
set -e
claude -p "Read the spec in docs/auth-spec.md. Implement OAuth2 login in src/auth/. Write tests first (TDD). Do NOT create any new documentation files."
claude -p "Review all files changed by the previous commit. Remove any unnecessary type tests, overly defensive checks, or testing of language features (e.g., testing that TypeScript generics work). Keep real business logic tests. Run the test suite after cleanup."
claude -p "Run the full build, lint, type check, and test suite. Fix any failures. Do not add new features."
claude -p "Create a conventional commit for all staged changes. Use 'feat: add OAuth2 login flow' as the message."
äž»èŠãªèšèšåå
åã¹ãããã¯ç¬ç« â claude -p ã³ãŒã«ããšã«æ°ããã³ã³ããã¹ããŠã£ã³ããŠãéå§ããããããã¹ãããéã§ã³ã³ããã¹ãã®æŒæŽ©ããããŸãããé åºãéèŠ â ã¹ãããã¯é 次å®è¡ãããŸããåã¹ãããã¯åã®ã¹ããããæ®ãããã¡ã€ã«ã·ã¹ãã ã®ç¶æ
ã«åºã¥ããŠåäœããŸããåŠå®çãªæ瀺ã¯å±éº â ãåã·ã¹ãã ããã¹ãããªãã§ããšã¯èšããªãã§ãã ããã代ããã«ãå¥ã®ã¯ãªãŒã³ã¢ããã¹ããããè¿œå ããŠãã ããïŒDe-Sloppify Pattern ãåç
§ïŒãçµäºã³ãŒãã®äŒæ¬ â set -e ã«ããã倱ææã«ãã€ãã©ã€ã³ãåæ¢ããŸãã
ããªãšãŒã·ã§ã³ ã¢ãã«ã«ãŒãã£ã³ã°ä»ã:
claude -p --model opus "Analyze the codebase architecture and write a plan for adding caching..."
claude -p "Implement the caching layer according to the plan in docs/caching-plan.md..."
claude -p --model opus "Review all changes for security issues, race conditions, and edge cases..."
ç°å¢ã³ã³ããã¹ãä»ã:
echo "Focus areas: auth module, API rate limiting" > .claude-context.md
claude -p "Read .claude-context.md for priorities. Work through them in order."
rm .claude-context.md
--allowedTools å¶éä»ã:
claude -p --allowedTools "Read,Grep,Glob" "Audit this codebase for security vulnerabilities..."
claude -p --allowedTools "Read,Write,Edit,Bash" "Implement the fixes from security-audit.md..."
2. NanoClaw REPL ECC çµã¿èŸŒã¿ã®æ°žç¶ã«ãŒãã§ãã å®å
šãªäŒè©±å±¥æŽä»ã㧠claude -p ãåæçã«åŒã³åºãã»ãã·ã§ã³å¯Ÿå¿ REPL ã§ãã
node scripts/claw.js
CLAW_SESSION=my-project CLAW_SKILLS=tdd-workflow,security-review node scripts/claw.js
ä»çµã¿
~/.claude/claw/{session}.md ããäŒè©±å±¥æŽãèªã¿èŸŒã¿ãŸãåãŠãŒã¶ãŒã¡ãã»ãŒãžã¯å®å
šãªå±¥æŽãã³ã³ããã¹ããšã㊠claude -p ã«éä¿¡ãããŸã
ã¬ã¹ãã³ã¹ã¯ã»ãã·ã§ã³ãã¡ã€ã«ã«è¿œèšãããŸãïŒMarkdown ãããŒã¿ããŒã¹ãšããŠå©çšïŒ
ã»ãã·ã§ã³ã¯åèµ·ååŸãæ°žç¶åãããŸã
NanoClaw vs Sequential Pipeline ã®äœ¿ãåã ãŠãŒã¹ã±ãŒã¹ NanoClaw Sequential Pipeline ã€ã³ã¿ã©ã¯ãã£ããªæ¢çŽ¢ ã¯ã ããã ã¹ã¯ãªããåãããèªåå ããã ã¯ã ã»ãã·ã§ã³æ°žç¶å çµã¿èŸŒã¿ æå ã³ã³ããã¹ãèç© ã¿ãŒã³ããšã«å¢å åã¹ãããã§æ°èŠ CI/CD çµ±å äžåã æé©
詳现㯠/claw ã³ãã³ãã®ããã¥ã¡ã³ããåç
§ããŠãã ããã
3. Infinite Agentic Loop 2ã€ã®ããã³ãããããªãã·ã¹ãã ã§ãä»æ§é§åã®çæã®ããã«äžŠåãµããšãŒãžã§ã³ãã調æŽããŸããdisler ãéçºïŒcredit: @dislerïŒã
ã¢ãŒããã¯ãã£: Two-Prompt System PROMPT 1 (Orchestrator) PROMPT 2 (Sub-Agents)
âââââââââââââââââââââââ ââââââââââââââââââââââââ
â ä»æ§ãã¡ã€ã«ãããŒã¹ â â ãã«ã³ã³ããã¹ããåä¿¡ â
â åºåãã£ã¬ã¯ããªãã¹ãã£ã³ â deploys â å²ãåœãŠçªå·ãèªã¿åã â
â ã€ãã¬ãŒã·ã§ã³ãèšç» ââââââââââââââ ä»æ§ã«æ£ç¢ºã«åŸã â
â ã¯ãªãšã€ãã£ããªæ¹åãå²åœâ N agents â ãŠããŒã¯ãªåºåãçæ â
â ãŠã§ãŒãã管ç â â åºåãã£ã¬ã¯ããªã«ä¿å â
âââââââââââââââââââââââ ââââââââââââââââââââââââ
ãã¿ãŒã³
ä»æ§åæ â ãªãŒã±ã¹ãã¬ãŒã¿ãŒãäœãçæããããå®çŸ©ããä»æ§ãã¡ã€ã«ïŒMarkdownïŒãèªã¿åããŸããã£ã¬ã¯ããªåµå¯ â æ¢åã®åºåãã¹ãã£ã³ããŠæ倧ã®ã€ãã¬ãŒã·ã§ã³çªå·ãèŠã€ããŸã䞊åããã〠â N åã®ãµããšãŒãžã§ã³ããèµ·åããããããã«ä»¥äžãä»äž:
å®å
šãªä»æ§
ãŠããŒã¯ãªã¯ãªãšã€ãã£ãã®æ¹åæ§
ç¹å®ã®ã€ãã¬ãŒã·ã§ã³çªå·ïŒç«¶åãªãïŒ
æ¢åã€ãã¬ãŒã·ã§ã³ã®ã¹ãããã·ã§ããïŒãŠããŒã¯æ§ã®ç¢ºä¿ã®ããïŒ
ãŠã§ãŒã管ç â ç¡éã¢ãŒãã§ã¯ãã³ã³ããã¹ããå°œãããŸã§3-5ãšãŒãžã§ã³ãã®ãŠã§ãŒãããããã€
å®è£
via Claude Code Commands .claude/commands/infinite.md ãäœæ:
Parse the following arguments from $ARGUMENTS:
1. spec_file â path to the specification markdown
2. output_ dir â where iterations are saved
3. count â integer 1-N or "infinite"
PHASE 1: Read and deeply understand the specification.
PHASE 2: List output_dir, find highest iteration number. Start at N+1.
PHASE 3: Plan creative directions â each agent gets a DIFFERENT theme/approach.
PHASE 4: Deploy sub-agents in parallel (Task tool). Each receives:
- Full spec text
- Current directory snapshot
- Their assigned iteration number
- Their unique creative direction
PHASE 5 (infinite mode): Loop in waves of 3-5 until context is low.
/project:infinite specs/component-spec.md src/ 5
/project:infinite specs/component-spec.md src/ infinite
ãããæŠç¥ ã«ãŠã³ã æŠç¥ 1-5 å
šãšãŒãžã§ã³ããåæå®è¡ 6-20 5åãã€ã®ããã infinite 3-5ã®ãŠã§ãŒãã段éçã«æŽç·ŽåºŠãåäž
éèŠãªæŽå¯: å²ãåœãŠã«ãããŠããŒã¯æ§ ãšãŒãžã§ã³ãèªèº«ã«å·®å¥åãä»»ããªãã§ãã ããããªãŒã±ã¹ãã¬ãŒã¿ãŒãåãšãŒãžã§ã³ãã«ç¹å®ã®ã¯ãªãšã€ãã£ãã®æ¹åæ§ãšã€ãã¬ãŒã·ã§ã³çªå·ã å²ãåœãŠ ãŸããããã«ããã䞊åãšãŒãžã§ã³ãéã§ã³ã³ã»ããã®éè€ãé²æ¢ãããŸãã
4. Continuous Claude PR Loop æ¬çªå質ã®ã·ã§ã«ã¹ã¯ãªãã ã§ãClaude Code ãç¶ç¶ã«ãŒãã§å®è¡ããPR ãäœæããCI ãåŸ
ã¡ãèªåçã«ããŒãžããŸããAnandChowdhary ãäœæïŒcredit: @AnandChowdharyïŒã
Core Loop âââââââââââââââââââââââââââââââââââââââââââââââââââââââ
â CONTINUOUS CLAUDE ITERATION â
â â
â 1. ãã©ã³ããäœæ (continuous-claude/iteration-N) â
â 2. æ¡åŒµããã³ããä»ã㧠claude -p ãå®è¡ â
â 3. (ãªãã·ã§ã³) ã¬ãã¥ãŒãã¹ â å¥ã® claude -p â
â 4. å€æŽãã³ãããïŒclaude ãã¡ãã»ãŒãžãçæïŒ â
â 5. ããã·ã¥ïŒPR äœæ (gh pr create) â
â 6. CI ãã§ãã¯ãåŸ
æ©ïŒgh pr checks ãããŒãªã³ã°ïŒ â
â 7. CI 倱æïŒâ èªåä¿®æ£ãã¹ (claude -p) â
â 8. PR ãããŒãž (squash/merge/rebase) â
â 9. main ã«æ»ã â ç¹°ãè¿ã â
â â
â å¶é: --max-runs N | --max-cost $X â
â --max-duration 2h | completion signal â
âââââââââââââââââââââââââââââââââââââââââââââââââââââââ
ã€ã³ã¹ããŒã«
èŠå: ã³ãŒãã確èªããäžã§ããªããžããªãã continuous-claude ãã€ã³ã¹ããŒã«ããŠãã ãããå€éšã¹ã¯ãªãããçŽæ¥ bash ã«ãã€ãããªãã§ãã ããã
䜿ãæ¹
continuous-claude --prompt "Add unit tests for all untested functions" --max-runs 10
continuous-claude --prompt "Fix all linter errors" --max-cost 5.00
continuous-claude --prompt "Improve test coverage" --max-duration 8h
continuous-claude \
--prompt "Add authentication feature" \
--max-runs 10 \
--review-prompt "Run npm test && npm run lint, fix any failures"
continuous-claude --prompt "Add tests" --max-runs 5 --worktree tests-worker &
continuous-claude --prompt "Refactor code" --max-runs 5 --worktree refactor-worker &
wait
ã€ãã¬ãŒã·ã§ã³éã®ã³ã³ããã¹ã: SHARED_TASK_NOTES.md éèŠãªã€ãããŒã·ã§ã³: SHARED_TASK_NOTES.md ãã¡ã€ã«ãã€ãã¬ãŒã·ã§ã³éã§æ°žç¶åãããŸã:
## Progress
- [x] Added tests for auth module (iteration 1)
- [x] Fixed edge case in token refresh (iteration 2)
- [ ] Still need: rate limiting tests, error boundary tests
## Next Steps
- Focus on rate limiting module next
- The mock setup in tests/helpers.ts can be reused
Claude ã¯ã€ãã¬ãŒã·ã§ã³éå§æã«ãã®ãã¡ã€ã«ãèªã¿åããã€ãã¬ãŒã·ã§ã³çµäºæã«æŽæ°ããŸããããã«ãããç¬ç«ãã claude -p åŒã³åºãéã®ã³ã³ããã¹ãã®ã£ãããæ©æž¡ããããŸãã
CI 倱æããã®å埩 PR ãã§ãã¯ã倱æããå ŽåãContinuous Claude ã¯èªåçã«ä»¥äžãè¡ããŸã:
gh run list çµç±ã§å€±æããã©ã³ ID ãååŸCI ä¿®æ£ã³ã³ããã¹ãä»ãã§æ°ãã claude -p ãèµ·å
Claude ã gh run view çµç±ã§ãã°ã確èªããã³ãŒããä¿®æ£ãã³ããããããã·ã¥
ãã§ãã¯ãååŸ
æ©ïŒæ倧 --ci-retry-max åã®è©Šè¡ïŒ
Completion Signal Claude ã¯ãå®äºããããžãã¯ãã¬ãŒãºã®åºåã§ã·ã°ãã«ã§ããŸã:
continuous-claude \
--prompt "Fix all bugs in the issue tracker" \
--completion-signal "CONTINUOUS_CLAUDE_PROJECT_COMPLETE" \
--completion-threshold 3
3åé£ç¶ã§å®äºã·ã°ãã«ãéä¿¡ãããšã«ãŒããåæ¢ããå®äºããäœæ¥ã«å¯Ÿããç¡é§ãªå®è¡ãé²ããŸãã
äž»èŠãªèšå® ãã©ã° ç®ç --max-runs NN åã®æåã€ãã¬ãŒã·ã§ã³åŸã«åæ¢ --max-cost $X$X ã®æ¯åºåŸã«åæ¢ --max-duration 2hçµéæéåŸã«åæ¢ --merge-strategy squashsquashãmergeããŸã㯠rebase --worktree <name>git worktree çµç±ã®äžŠåå®è¡ --disable-commitsãã©ã€ã©ã³ã¢ãŒãïŒgit æäœãªãïŒ --review-prompt "..."ã€ãã¬ãŒã·ã§ã³ããšã«ã¬ãã¥ãŒãã¹ãè¿œå --ci-retry-max NCI 倱æã®èªåä¿®æ£ïŒããã©ã«ã: 1ïŒ
5. The De-Sloppify Pattern ä»»æã®ã«ãŒãã®ã¢ããªã³ãã¿ãŒã³ã§ãã Implementer ã¹ãããã®åŸã«å°çšã®ã¯ãªãŒã³ã¢ãã/ãªãã¡ã¯ã¿ãªã³ã°ã¹ããããè¿œå ããŸãã
åé¡ LLM ã« TDD ã§å®è£
ãäŸé Œãããšãããã¹ããæžãããæåéãã«è§£éããããŸã:
TypeScript ã®åã·ã¹ãã ãåäœããããšãæ€èšŒãããã¹ãïŒtypeof x === 'string' ã®ãã¹ãïŒ
åã·ã¹ãã ãæ¢ã«ä¿èšŒããŠãããã®ã«å¯Ÿããé床ã«é²åŸ¡çãªã©ã³ã¿ã€ã ãã§ãã¯
ããžãã¹ããžãã¯ã§ã¯ãªããã¬ãŒã ã¯ãŒã¯ã®åäœããã¹ã
å®éã®ã³ãŒããææ§ã«ããéå°ãªãšã©ãŒãã³ããªã³ã°
ãªãåŠå®çãªæ瀺ã§ã¯ãã¡ãïŒ Implementer ããã³ããã«ãåã·ã¹ãã ããã¹ãããªãã§ãããäžèŠãªãã§ãã¯ãè¿œå ããªãã§ããè¿œå ãããšãäžæµã«åœ±é¿ãåºãŸã:
ã¢ãã«ããã¹ãŠã®ãã¹ãã«å¯ŸããŠæ¶æ¥µçã«ãªã
æ£åœãªãšããžã±ãŒã¹ãã¹ããã¹ããããã
å質ãäºæž¬äžèœã«äœäžãã
解決ç: å¥ã®ãã¹ Implementer ãå¶çŽãã代ããã«ã培åºçã«äœæ¥ãããŸãããã®åŸãéäžããã¯ãªãŒã³ã¢ãããšãŒãžã§ã³ããè¿œå ããŸã:
claude -p "Implement the feature with full TDD. Be thorough with tests."
claude -p "Review all changes in the working tree. Remove:
- Tests that verify language/framework behavior rather than business logic
- Redundant type checks that the type system already enforces
- Over-defensive error handling for impossible states
- Console.log statements
- Commented-out code
Keep all business logic tests. Run the test suite after cleanup to ensure nothing breaks."
ã«ãŒãã³ã³ããã¹ãã§ã®äœ¿çš for feature in "${features[@]} " ; do
claude -p "Implement $feature with TDD."
claude -p "Cleanup pass: review changes, remove test/code slop, run tests."
claude -p "Run build + lint + tests. Fix any failures."
claude -p "Commit with message: feat: add $feature "
done
éèŠãªæŽå¯
äžæµã®å質ã«åœ±é¿ãäžããåŠå®çãªæ瀺ãè¿œå ãã代ããã«ãå¥ã® de-sloppify ãã¹ãè¿œå ããŠãã ãããéäžãã2ã€ã®ãšãŒãžã§ã³ãã¯ãå¶çŽããã1ã€ã®ãšãŒãžã§ã³ããããåªããŠããŸãã
6. Ralphinho / RFC-Driven DAG Orchestration æãé«åºŠãªãã¿ãŒã³ã§ãã RFC é§åã®ãã«ããšãŒãžã§ã³ããã€ãã©ã€ã³ã§ãä»æ§ãäŸåé¢ä¿ DAG ã«å解ããåãŠããããéå±€çãªå質ãã€ãã©ã€ã³ãéããŠå®è¡ãããšãŒãžã§ã³ãé§åã®ããŒãžãã¥ãŒãä»ããŠã©ã³ãã£ã³ã°ããŸããenitrat ãäœæïŒcredit: @enitratïŒã
ã¢ãŒããã¯ã㣠Overview RFC/PRD Document
â
âŒ
DECOMPOSITION (AI)
RFC ãã¯ãŒã¯ãŠããããšäŸåé¢ä¿ DAG ã«å解
â
âŒ
ââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
â RALPH LOOP (æ倧3ãã¹) â
â â
â å DAG ã¬ã€ã€ãŒã«ã€ããŠïŒäŸåé¢ä¿é ã«é次ïŒ: â
â â
â âââ Quality PipelinesïŒãŠãããããšã«äžŠåïŒâââââââââ â
â â åãŠãããã¯ç¬èªã® worktree å
: â â
â â Research â Plan â Implement â Test â Review â â
â â ïŒæ·±åºŠã¯è€é床ãã£ã¢ã«ããå€åïŒ â â
â ââââââââââââââââââââââââââââââââââââââââââââââââââ â
â â
â âââ Merge Queue ââââââââââââââââââââââââââââââââââ â
â â main ã«ãªããŒã¹ â ãã¹ãå®è¡ â ã©ã³ããŸãã¯è¿œæŸâ â
â â è¿œæŸããããŠãããã¯ã³ã³ããªã¯ãã³ã³ããã¹ãä»ãã§â â
â â åæå
¥ â â
â ââââââââââââââââââââââââââââââââââââââââââââââââââ â
â â
ââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
RFC å解 AI ã RFC ãèªã¿åããã¯ãŒã¯ãŠããããçæããŸã:
interface WorkUnit {
id : string ;
name : string ;
rfcSections : string [];
description : string ;
deps : string [];
acceptance : string [];
tier : "trivial" | "small" | "medium" | "large" ;
}
å°æ°ã®åéæ§ã®ãããŠããããåªå
ïŒããŒãžãªã¹ã¯ã®æå°åïŒ
ãŠãããéã®ãã¡ã€ã«éè€ãæå°åïŒã³ã³ããªã¯ãåé¿ïŒ
ãã¹ãã¯å®è£
ãšäžç·ã«ä¿æïŒãX ãå®è£
ããšãX ããã¹ãããåé¢ããªãïŒ
å®éã®ã³ãŒãäŸåé¢ä¿ãããå Žåã®ã¿äŸåé¢ä¿ãèšå®
äŸåé¢ä¿ DAG ãå®è¡é åºã決å®ããŸã:
Layer 0: [unit-a, unit-b] â äŸåé¢ä¿ãªãã䞊åå®è¡
Layer 1: [unit-c] â unit-a ã«äŸå
Layer 2: [unit-d, unit-e] â unit-c ã«äŸå
è€é床ãã£ã¢ ç°ãªããã£ã¢ã«ç°ãªããã€ãã©ã€ã³æ·±åºŠãé©çšãããŸã:
ãã£ã¢ ãã€ãã©ã€ã³ã¹ããŒãž trivial implement â test small implement â test â code-review medium research â plan â implement â test â PRD-review + code-review â review-fix large research â plan â implement â test â PRD-review + code-review â review-fix â final-review
ããã«ãããã·ã³ãã«ãªå€æŽã«å¯Ÿããé«ã³ã¹ããªæäœãé²ãã€ã€ãã¢ãŒããã¯ãã£ã®å€æŽã«ã¯åŸ¹åºçãªç²Ÿæ»ãè¡ãããŸãã
ç¬ç«ããã³ã³ããã¹ããŠã£ã³ããŠïŒèè
ãã€ã¢ã¹ã®æé€ïŒ åã¹ããŒãžã¯ç¬èªã®ãšãŒãžã§ã³ãããã»ã¹ãšç¬èªã®ã³ã³ããã¹ããŠã£ã³ããŠã§å®è¡ãããŸã:
ã¹ããŒãž ã¢ãã« ç®ç Research Sonnet ã³ãŒãããŒã¹ïŒRFC ãèªã¿åããã³ã³ããã¹ãããã¥ã¡ã³ããçæ Plan Opus å®è£
ã¹ããããèšèš Implement Codex ãã©ã³ã«åŸã£ãŠã³ãŒããæžã Test Sonnet ãã«ãïŒãã¹ãã¹ã€ãŒããå®è¡ PRD Review Sonnet ä»æ§æºæ ãã§ã㯠Code Review Opus å質ïŒã»ãã¥ãªãã£ãã§ã㯠Review Fix Codex ã¬ãã¥ãŒææãžã®å¯Ÿå¿ Final Review Opus å質ã²ãŒãïŒlarge ãã£ã¢ã®ã¿ïŒ
éèŠãªèšèš: ã¬ãã¥ãŒã¢ãŒãã¬ãã¥ãŒããã³ãŒããæžããããšã¯ãããŸãããããã«ããèè
ãã€ã¢ã¹ãæé€ãããŸããããã¯ã»ã«ãã¬ãã¥ãŒã§èŠéãããåé¡ã®æãäžè¬çãªåå ã§ãã
è¿œæŸä»ãããŒãžãã¥ãŒ å質ãã€ãã©ã€ã³å®äºåŸããŠãããã¯ããŒãžãã¥ãŒã«å
¥ããŸã:
Unit branch
â
ââ main ã«ãªããŒã¹
â ââ ã³ã³ããªã¯ãïŒâ è¿œæŸïŒã³ã³ããªã¯ãã³ã³ããã¹ãããã£ããã£ïŒ
â
ââ ãã«ãïŒãã¹ãå®è¡
â ââ 倱æïŒâ è¿œæŸïŒãã¹ãåºåããã£ããã£ïŒ
â
ââ ãã¹ â main ã fast-forwardãããã·ã¥ããã©ã³ãåé€
ãã¡ã€ã«éè€ã®ã€ã³ããªãžã§ã³ã¹:
éè€ã®ãªããŠãããã¯ææ©çã«äžŠåã§ã©ã³ãã£ã³ã°
éè€ã®ãããŠãããã¯1ã€ãã€ãæ¯åãªããŒã¹ããªããã©ã³ãã£ã³ã°
è¿œæŸããã®å埩:
è¿œæŸæãå®å
šãªã³ã³ããã¹ãïŒã³ã³ããªã¯ããã¡ã€ã«ãdiffããã¹ãåºåïŒããã£ããã£ããã次㮠Ralph ãã¹ã§å®è£
è
ã«ãã£ãŒãããã¯ãããŸã:
## MERGE CONFLICT â RESOLVE BEFORE NEXT LANDING
ããªãã®åã®å®è£
ã¯ãå
ã«ã©ã³ãã£ã³ã°ããå¥ã®ãŠããããšã³ã³ããªã¯ãããŸããã
以äžã®ã³ã³ããªã¯ãããŠãããã¡ã€ã«/è¡ãåé¿ãããããå€æŽãåæ§ç¯ããŠãã ããã
{full eviction context with diffs}
ã¹ããŒãžéã®ããŒã¿ãã㌠research.contextFilePath âââââââââââââââââââ plan
plan.implementationSteps âââââââââââââââââââ implement
implement.{filesCreated, whatWasDone} ââââââ test, reviews
test.failingSummary ââââââââââââââââââââââââ reviews, implement (next pass)
reviews.{feedback, issues} âââââââââââââââââ review-fix â implement (next pass)
final-review.reasoning âââââââââââââââââââââ implement (next pass)
evictionContext ââââââââââââââââââââââââââââ implement (after merge conflict)
Worktree ã«ããéé¢ ãã¹ãŠã®ãŠãããã¯éé¢ããã worktree ã§å®è¡ãããŸãïŒgit ã§ã¯ãªã jj/Jujutsu ã䜿çšïŒ:
/tmp/workflow-wt-{unit-id}/
åããŠãããã®ãã€ãã©ã€ã³ã¹ããŒãžã¯ worktree ã å
±æ ããresearch â plan â implement â test â review ã«ããã£ãŠç¶æ
ïŒã³ã³ããã¹ããã¡ã€ã«ããã©ã³ãã¡ã€ã«ãã³ãŒãå€æŽïŒãä¿æããŸãã
äž»èŠãªèšèšåå
決å®è«çãªå®è¡ â äºåã®å解ã«ãã䞊åæ§ãšé åºã確å®ã¬ãã¬ããžãã€ã³ãã§ã®äººéã¬ãã¥ãŒ â äœæ¥èšç»ãå¯äžæ倧ã®ã¬ãã¬ããžãšãªãä»å
¥ãã€ã³ãé¢å¿ã®åé¢ â åã¹ããŒãžã¯å¥ã®ã³ã³ããã¹ããŠã£ã³ããŠã§å¥ã®ãšãŒãžã§ã³ããšããŠå®è¡ã³ã³ããã¹ãä»ãã®ã³ã³ããªã¯ãå埩 â å®å
šãªè¿œæŸã³ã³ããã¹ãã«ãããç²ç®çãªãªãã©ã€ã§ã¯ãªãã€ã³ããªãžã§ã³ããªåå®è¡ãå¯èœãã£ã¢é§åã®æ·±åºŠ â äºçŽ°ãªå€æŽã¯ research/review ãã¹ãããã倧èŠæš¡ãªå€æŽã¯æ倧ã®ç²Ÿæ»åéå¯èœãªã¯ãŒã¯ãã㌠â å®å
šãªç¶æ
ã SQLite ã«æ°žç¶åãä»»æã®ãã€ã³ãããåéå¯èœ
䜿çšã¿ã€ãã³ã° Ralphinho vs Simpler Patterns ã·ã°ãã« Ralphinho ãäœ¿çš ã·ã³ãã«ãªãã¿ãŒã³ãäœ¿çš è€æ°ã®çžäºäŸåããã¯ãŒã¯ãŠããã ã¯ã ããã 䞊åå®è£
ãå¿
èŠ ã¯ã ããã ããŒãžã³ã³ããªã¯ãã®å¯èœæ§ ã¯ã ãããïŒé次ã§ååïŒ åäžãã¡ã€ã«ã®å€æŽ ããã ã¯ãïŒsequential pipelineïŒ è€æ°æ¥ã®ãããžã§ã¯ã ã¯ã å Žåã«ããïŒcontinuous-claudeïŒ ä»æ§/RFC ãæ¢ã«æžãããŠãã ã¯ã å Žåã«ãã 1ã€ã®ããšã«å¯ŸããçŽ æ©ãã€ãã¬ãŒã·ã§ã³ ããã ã¯ãïŒNanoClaw ãŸãã¯ãã€ãã©ã€ã³ïŒ
é©åãªãã¿ãŒã³ã®éžæ
æ±ºå® Matrix ã¿ã¹ã¯ã¯åäžã®éäžããå€æŽãïŒ
ââ ã¯ã â Sequential Pipeline ãŸã㯠NanoClaw
ââ ããã â æžãããä»æ§/RFC ããããïŒ
ââ ã¯ã â 䞊åå®è£
ãå¿
èŠãïŒ
â ââ ã¯ã â RalphinhoïŒDAG ãªãŒã±ã¹ãã¬ãŒã·ã§ã³ïŒ
â ââ ããã â Continuous ClaudeïŒå埩 PR ã«ãŒãïŒ
ââ ããã â åããã®ã®ããªãšãŒã·ã§ã³ãå€æ°å¿
èŠãïŒ
ââ ã¯ã â Infinite Agentic LoopïŒä»æ§é§åã®çæïŒ
ââ ããã â Sequential Pipeline + de-sloppify
ãã¿ãŒã³ã®çµã¿åãã ãããã®ãã¿ãŒã³ã¯ããŸãçµã¿åãããããŸã:
Sequential Pipeline + De-Sloppify â æãäžè¬çãªçµã¿åããã§ãããã¹ãŠã®å®è£
ã¹ãããã«ã¯ãªãŒã³ã¢ãããã¹ãä»ããŸãã
Continuous Claude + De-Sloppify â åã€ãã¬ãŒã·ã§ã³ã« de-sloppify ãã£ã¬ã¯ãã£ãä»ãã® --review-prompt ãè¿œå ããŸãã
ä»»æã®ã«ãŒã + Verification â ã³ãããåã®ã²ãŒããšã㊠ECC ã® /verify ã³ãã³ããŸã㯠verification-loop ã¹ãã«ã䜿çšããŸãã
Ralphinho ã®ãã£ã¢ã¢ãããŒããã·ã³ãã«ãªã«ãŒã㧠â sequential pipeline ã§ããã·ã³ãã«ãªã¿ã¹ã¯ã Haiku ã«ãè€éãªã¿ã¹ã¯ã Opus ã«ã«ãŒãã£ã³ã°ã§ããŸã:
claude -p --model haiku "Fix the import ordering in src/utils.ts"
claude -p --model opus "Refactor the auth module to use the strategy pattern"
ã¢ã³ããã¿ãŒã³
ããããééã
çµäºæ¡ä»¶ã®ãªãç¡éã«ãŒã â åžžã« max-runsãmax-costãmax-durationããŸã㯠completion signal ãèšå®ããŠãã ããã
ã€ãã¬ãŒã·ã§ã³éã®ã³ã³ããã¹ãæ©æž¡ããªã â å claude -p ã³ãŒã«ã¯æ°èŠã§å§ãŸããŸããSHARED_TASK_NOTES.md ããã¡ã€ã«ã·ã¹ãã ã®ç¶æ
ã䜿ã£ãŠã³ã³ããã¹ããæ©æž¡ãããŠãã ããã
åã倱æã®ãªãã©ã€ â ã€ãã¬ãŒã·ã§ã³ã倱æããå Žåãåã«ãªãã©ã€ããªãã§ãã ããããšã©ãŒã³ã³ããã¹ãããã£ããã£ããŠæ¬¡ã®è©Šè¡ã«ãã£ãŒãããŠãã ããã
ã¯ãªãŒã³ã¢ãããã¹ã®ä»£ããã«åŠå®çãªæ瀺 â ãX ãããªãã§ããšèšããªãã§ãã ãããX ãé€å»ããå¥ã®ãã¹ãè¿œå ããŠãã ããã
ãã¹ãŠã®ãšãŒãžã§ã³ãã1ã€ã®ã³ã³ããã¹ããŠã£ã³ããŠã« â è€éãªã¯ãŒã¯ãããŒã§ã¯ãé¢å¿ãç°ãªããšãŒãžã§ã³ãããã»ã¹ã«åé¢ããŠãã ãããã¬ãã¥ãŒã¢ãŒã¯èè
ã§ãã£ãŠã¯ãªããŸããã
䞊åäœæ¥ã§ã®ãã¡ã€ã«éè€ã®ç¡èŠ â 2ã€ã®äžŠåãšãŒãžã§ã³ããåããã¡ã€ã«ãç·šéããå¯èœæ§ãããå ŽåãããŒãžæŠç¥ïŒé次ã©ã³ãã£ã³ã°ããªããŒã¹ããŸãã¯ã³ã³ããªã¯ã解決ïŒãå¿
èŠã§ãã
References ãããžã§ã¯ã äœè
ãªã³ã¯ Ralphinho enitrat credit: @enitrat Infinite Agentic Loop disler credit: @disler Continuous Claude AnandChowdhary credit: @AnandChowdhary NanoClaw ECC ãã®ãªããžããªã® /claw ã³ãã³ã Verification Loop ECC ãã®ãªããžããªã® skills/verification-loop/
