Autonomous development loop patterns for iterative self-improvement. Auto-triggers when implementing features autonomously, fixing bugs in a loop, or running until completion. Based on frankbria/ralph-claude-code.
SkillScore72Maintained
On-demandOn-demand: invoked explicitly — never occupies your agent's context until called.
◆Coding pack
Last verified 8/19/2026
What it changes
Passed 14 of 20 test cases with the skill, and 3 of 20 without it.
Without skill16%
With skill72%
Checks it passed that the model alone missed
Sets initial iteration field to 0
Updates phase field to waiting-user
Sets initial phase field to analyze
and 15 more in the full report
15 fixed3 passed either way6 still failing1 broke
Pass-rate lift
+56 pp
Tokens
+39%
Agent turns
0%
Response time
-70%
Model gemini-3.6-flashMeasured 8/19/2026v218 of 25 cases passed with skill · 20 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 15 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 15
Prompt
An autonomous task paused for user feedback via Slack notification. If 10 minutes pass without a user response, what exact value is written to the phase field in ralph-state.md?
What was checked, without and with the skill
Checked
Without skill
With skill
Updates phase field to waiting-user
Missed
Passed
case-16 · gemini-3.6-flash · 8/19/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md148 linessingle file
Autonomous Loop
Plan→Execute→Verify를 반복하여 복잡한 작업을 자율적으로 완성합니다.
사용자 개입 없이 계획 수립부터 구현, 검증, 수정까지 수행합니다.
Task-Folder 구조
모든 상태 파일은 작업별 폴더에 격리됩니다:
.claude/project/state/{ISO8601-basic}_{task-name}/
├── ralph-state.md # 루프 제어 메타데이터
├── notepad.md # 작업 메모 (note 스킬)
└── prd.json # PRD (선택)
폴더 네이밍 예시: 2026-02-13T1430_login-feature
모드
모드
설명
최대 반복
병렬 실행
Ralph
완료 보장. 상태 파일 추적, PRD 연동, 완료까지 반복
10 (설정 가능)
선택적
Autopilot
최대 성능. 병렬 에이전트, 빠른 반복
3
적극적
Ralph: 사용자가 "끝까지", "완료할 때까지" 등을 요청할 때
Autopilot: 사용자가 "최대 성능", "병렬로" 등을 요청할 때
참조 스킬
Read tool로 다음 스킬을 읽어 적용하세요:
.claude/skills/verify-loop/SKILL.md — 자기 수정 루프, 4-Level 완료 기준
.claude/skills/planning/SKILL.md — 개발 계획 수립
.claude/skills/code-accuracy/SKILL.md — 코드 정확성 검증
.claude/skills/notify-user/SKILL.md — 에스컬레이션 시 Slack-First 모드로 사용자 응답 수신
project CLAUDE.md or .claude/rules/ — 프로젝트별 패턴 및 컨벤션
ALWAYS activate when user types /autoresearch, $autoresearch plan, $autoresearch debug, $autoresearch fix, $autoresearch security, $autoresearch ship, $autoresearch scenario, $autoresearch predict, $autoresearch learn, $autoresearch reason, or $autoresearch probe. MUST also activate when user mentions "autoresearch" with ANY goal, metric, or task, even when the invocation is embedded in prose. This is a BLOCKING skill invocation — invoke BEFORE generating any other response.