DaeSeokSong/oh-my-ouroboros
🐍 Like the mythic serpent coiled in a ring, jaws closed on its own tail — this harness devours its own output to grow: monthly self-audits, feedback loops, token-budget discipline. 25 skills · 10 agents · 40 lazy rules, mirrored from a prod
stack: Python
ARCHETYPE
YOLO Cowboy
Permissions? Never heard of 'em. Ships at the speed of `--dangerously-skip-permissions`.
GUARDRAILS
2/5
Blocks destructive commands · Pre-tool screening hook · details
Quality 2/100 Guardrails v2
Copy this rig
# review before running: this installs third-party code $ npx degit DaeSeokSong/oh-my-ouroboros/agents ./rig-oh-my-ouroboros/agents $ npx degit DaeSeokSong/oh-my-ouroboros/skills ./rig-oh-my-ouroboros/skills
MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.
$ npx degit DaeSeokSong/oh-my-ouroboros/skills/ckpt-cleanup .claude/skills/ckpt-cleanup $ npx degit DaeSeokSong/oh-my-ouroboros/skills/commit .claude/skills/commit $ npx degit DaeSeokSong/oh-my-ouroboros/skills/confluence-research-record .claude/skills/confluence-research-record $ npx degit DaeSeokSong/oh-my-ouroboros/skills/confluence .claude/skills/confluence $ npx degit DaeSeokSong/oh-my-ouroboros/skills/cost-reset .claude/skills/cost-reset $ npx degit DaeSeokSong/oh-my-ouroboros/skills/cost .claude/skills/cost $ npx degit DaeSeokSong/oh-my-ouroboros/skills/discuss .claude/skills/discuss $ npx degit DaeSeokSong/oh-my-ouroboros/skills/drawio-skill .claude/skills/drawio-skill $ npx degit DaeSeokSong/oh-my-ouroboros/skills/dumbify .claude/skills/dumbify $ npx degit DaeSeokSong/oh-my-ouroboros/skills/experiment .claude/skills/experiment $ npx degit DaeSeokSong/oh-my-ouroboros/skills/experimental-design .claude/skills/experimental-design $ npx degit DaeSeokSong/oh-my-ouroboros/skills/figure .claude/skills/figure $ npx degit DaeSeokSong/oh-my-ouroboros/skills/graphify .claude/skills/graphify $ npx degit DaeSeokSong/oh-my-ouroboros/skills/grok .claude/skills/grok $ npx degit DaeSeokSong/oh-my-ouroboros/skills/harness-audit .claude/skills/harness-audit $ npx degit DaeSeokSong/oh-my-ouroboros/skills/harness-bench .claude/skills/harness-bench $ npx degit DaeSeokSong/oh-my-ouroboros/skills/intent .claude/skills/intent $ npx degit DaeSeokSong/oh-my-ouroboros/skills/learn-from-feedback .claude/skills/learn-from-feedback $ npx degit DaeSeokSong/oh-my-ouroboros/skills/neuro-explain .claude/skills/neuro-explain $ npx degit DaeSeokSong/oh-my-ouroboros/skills/paper-read .claude/skills/paper-read $ npx degit DaeSeokSong/oh-my-ouroboros/skills/paper .claude/skills/paper $ npx degit DaeSeokSong/oh-my-ouroboros/skills/project-init .claude/skills/project-init $ npx degit DaeSeokSong/oh-my-ouroboros/skills/pydicom .claude/skills/pydicom $ npx degit DaeSeokSong/oh-my-ouroboros/skills/quick-skill .claude/skills/quick-skill $ npx degit DaeSeokSong/oh-my-ouroboros/skills/research .claude/skills/research $ npx degit DaeSeokSong/oh-my-ouroboros/skills/setup-panels .claude/skills/setup-panels $ npx degit DaeSeokSong/oh-my-ouroboros/skills/solve .claude/skills/solve $ npx degit DaeSeokSong/oh-my-ouroboros/skills/statistical-analysis .claude/skills/statistical-analysis $ npx degit DaeSeokSong/oh-my-ouroboros/skills/statistical-power .claude/skills/statistical-power $ npx degit DaeSeokSong/oh-my-ouroboros/skills/storytelling .claude/skills/storytelling $ npx degit DaeSeokSong/oh-my-ouroboros/skills/wandb-cleanup .claude/skills/wandb-cleanup $ npx degit DaeSeokSong/oh-my-ouroboros/skills/workflow .claude/skills/workflow
$ curl -fsSL --create-dirs -o .claude/agents/citation-verifier.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/citation-verifier.md $ curl -fsSL --create-dirs -o .claude/agents/coder.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/coder.md $ curl -fsSL --create-dirs -o .claude/agents/designer.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/designer.md $ curl -fsSL --create-dirs -o .claude/agents/docs.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/docs.md $ curl -fsSL --create-dirs -o .claude/agents/gitops.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/gitops.md $ curl -fsSL --create-dirs -o .claude/agents/planner.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/planner.md $ curl -fsSL --create-dirs -o .claude/agents/qa.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/qa.md $ curl -fsSL --create-dirs -o .claude/agents/researcher.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/researcher.md $ curl -fsSL --create-dirs -o .claude/agents/reviewer.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/reviewer.md $ curl -fsSL --create-dirs -o .claude/agents/scout.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/scout.md $ curl -fsSL --create-dirs -o .claude/agents/searcher.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/searcher.md $ curl -fsSL --create-dirs -o .claude/agents/security.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/security.md $ curl -fsSL --create-dirs -o .claude/agents/tutor.md https://raw.githubusercontent.com/DaeSeokSong/oh-my-ouroboros/main/agents/tutor.md
Merge into .claude/settings.json (project) or ~/.claude/settings.json (user). Hook commands reference scripts in the source repo — copy those too.
{
"permissions": {
"deny": [
"Bash(sudo:*)",
"Bash(shutdown:*)",
"Bash(reboot:*)",
"Read(./outputs/**)",
"Read(./lightning_logs/**)",
"Read(./wandb/**)",
"Read(./mlruns/**)",
"Read(./checkpoints/**)",
"Read(**/__pycache__/**)",
"Read(./node_modules/**)",
"Read(./dist/**)",
"Read(./build/**)",
"Read(**/*.ckpt)",
"Read(**/*.pth)",
"Read(**/*.safetensors)",
"Read(./_bench/**)",
"Read(./_cache/**)",
"Read(./_refs/**)"
]
},
"hooks": {
"PreToolUse": [
{
"matcher": "Write|Edit",
"hooks": [
{
"type": "command",
"command": "INPUT=$(cat); FILE=$(echo \"$INPUT\" | jq -r '.tool_input.file_path // .tool_input.path // empty' 2>/dev/null); if [ -n \"$FILE\" ] && echo \"$FILE\" | grep -qE '(\\.env$|\\.env\\.|(^|\\/)\\.?credentials(\\.json)?$|\\.pem$|\\.key$|\\.p12$|\\.pfx$|\\/secrets"
},
{
"type": "command",
"command": "bash ~/.claude/scripts/scope-gate.sh"
}
]
},
{
"matcher": "Bash",
"hooks": [
{
"type": "command",
"command": "bash ~/.claude/scripts/bash-write-check.sh"
},
{
"type": "command",
"command": "bash ~/.claude/scripts/git-guardrails.sh"
},
{
"type": "command",
"command": "bash $HOME/.claude/scripts/train-launch-warn.sh"
},
{
"type": "command",
"command": "python3 $HOME/.claude/scripts/external-model-gate.py"
}
]
},
{
"matcher": "Read",
"hooks": [
{
"type": "command",
"command": "$HOME/.claude/scripts/read-once.sh"
}
]
},
{
"matcher": "Agent",
"hooks": [
{
"type": "command",
"command": "bash $HOME/.claude/scripts/agent-spawn-log.sh pre"
}
]
},
{
"matcher": "mcp__claude_ai_Atlassian__(update|create)ConfluencePage",
"hooks": [
{
"type": "command",
"command": "bash $HOME/.claude/scripts/korean-style-gate.sh"
}
]
}
]
}
} Skills (32)
ckpt-cleanupcommitconfluence-research-recordconfluencecost-resetcostdiscussdrawio-skilldumbifyexperimentexperimental-designfiguregraphifygrokharness-auditharness-benchintentlearn-from-feedbackneuro-explainpaper-readpaperproject-initpydicomquick-skillresearchsetup-panelssolvestatistical-analysisstatistical-powerstorytellingwandb-cleanupworkflow
Subagents (13)
| citation-verifier model: sonnet | 논문 인용 신뢰도 검증 전문 에이전트 — 철회/정정 감지 + 학계 stance(지지/반박) 디베이트 분류. 논문을 본문·실험·근거로 인용하기 직전, SOTA 베이스라인·신경과학 메커니즘 참조 시 사용. scite 대체(무료 API + 전문 직접 읽기). 옹호/회의/종합/심사평수집 4개 모드로 호출되어 적대적 검증 수행. |
| coder model: sonnet | 코드 작성·수정·리뷰 전문 에이전트. scout이 정찰한 결과를 받아 정밀 구현. 코드 수정/작성 요청 시 사용. |
| designer model: sonnet | Figure 생성·TikZ 다이어그램·matplotlib 플롯·시각적 디자인 전문 에이전트. Figure/다이어그램/시각화 작업 시 사용. Figure 스크립트 직접 실행 허용. |
| docs model: sonnet | Confluence·Jira 문서 작업 전문 에이전트. 페이지 생성·수정, 이슈 관리 시 사용. MCP(confluence) 접근 권한 보유. git 작업은 gitops에게 위임. |
| gitops model: sonnet | git 전문 에이전트. 커밋·브랜치·diff·log·stash·rebase·merge 등 git 작업 시 사용. push/pull은 사용자 직접 수행 — gitops는 명령어만 안내. |
| planner model: opus | 아키텍처 설계, 실험 설계, 워크플로우 계획 전문 에이전트. 설계·기획 요청, 실험 순서 결정, 워크플로우 구축 시 사용. 계획 설계만 담당 — 코드 수정/생성 불가. |
| qa model: sonnet | 코드 검증·테스트·호출부 영향 분석 전문 에이전트. 구조적·수학적 정확성을 최우선 검증. 코드 수정 완료 후 scout QA정찰 결과를 받아 검증 실행. ML 실험 사전 체크 포함. |
| researcher model: opus | 논문 작성·실험 결과 분석·실험 설계 비판적 평가 전문 에이전트. 논문/연구 작업, 실험 결과 해석, 학술 콘텐츠 생성, planner 설계 타당성 평가 시 사용. |
| reviewer model: opus | 한 함수씩 파고드는 DFS 코드 리뷰 전문 에이전트. 설계 의도와 코드의 어긋남(근거 없는 상수·주석과 코드 불일치·도달 불가 갈래·이름과 동작 불일치·설계 문서와의 차이)을 찾는다. 브랜치·모듈 전수 리뷰, "코드 리뷰해줘"·"파고들어" 발화에 사용. 동작 검증은 qa, 취약점은 security 소관이라 겹치지 않는다. |
| scout model: haiku | 코드베이스·로컬 파일 정찰 전문 에이전트(coder·qa의 조수). "파일 찾아", "코드 어디 있어", "어떤 파일 봐야 해" 등 탐색 요청 시 사용. 코드 수정/생성 불가 — 탐색과 보고만 수행. |
| searcher model: sonnet | 인터넷 검색 전문 에이전트. "검색해", "찾아봐", "논문 조사", "라이브러리 문서", "최신 동향" 등 인터넷 기반 자료 수집 요청 시 사용. 코드베이스 탐색은 scout에게 위임. |
| security model: sonnet | 보안 취약점 검토 전문 에이전트 — 4관점(입력 주입·인증/인가·시크릿 노출·의존성/공급망) 코드 보안 리뷰. 보안 리뷰/취약점 점검/injection·인증 우회 의심 시 사용. 대규모 스캔은 claude-security 플러그인(main 경유, 깊이 게이트 필수). |
| tutor model: opus | 사용자의 완벽한 이해를 돕는 선생 에이전트. 논문·리서치 리포트·실험 결과 등 이미 만들어진 산출물을 재료로 출제→오답 진단→난도 하강 재설명→재출제를 전 문항 정답까지 반복. '이해 안 돼', '설명해줘', '퀴즈 내줘', '가르쳐줘' 발화와 /paper-read Step 7·/research 종료 단계에서 사용. |
Hooks (30)
| event | matcher | runs |
|---|---|---|
| PreToolUse | Write|Edit | INPUT=$(cat); FILE=$(echo "$INPUT" | jq -r '.tool_input.file_path // .tool_input.path // empty' 2>/dev/null); if [ -n "$FILE" ] && echo "$FILE" | grep -qE '(\.env$|\.env\.|(^|\/)\.?credentials(\.json)?$|\.pem$|\.key$|\.p12$|\.pfx$|\/secrets |
| PreToolUse | Write|Edit | bash ~/.claude/scripts/scope-gate.sh |
| PreToolUse | Bash | bash ~/.claude/scripts/bash-write-check.sh |
| PreToolUse | Bash | bash ~/.claude/scripts/git-guardrails.sh |
| PreToolUse | Bash | bash $HOME/.claude/scripts/train-launch-warn.sh |
| PreToolUse | Bash | python3 $HOME/.claude/scripts/external-model-gate.py |
| PreToolUse | Read | $HOME/.claude/scripts/read-once.sh |
| PreToolUse | Agent | bash $HOME/.claude/scripts/agent-spawn-log.sh pre |
| PreToolUse | mcp__claude_ai_Atlassian__(update|create)ConfluencePage | bash $HOME/.claude/scripts/korean-style-gate.sh |
| PostToolUse | Agent | bash $HOME/.claude/scripts/agent-spawn-log.sh post |
| PostToolUse | Write|Edit | $HOME/.claude/scripts/ci-gate-edit.sh |
| PostToolUse | Write|Edit | bash $HOME/.claude/scripts/graphify-autoindex.sh || true |
| PostToolUse | Read | bash $HOME/.claude/scripts/read-once.sh --record |
| PostToolUse | Bash | bash $HOME/.claude/scripts/graphify-commit-refresh.sh |
| Notification | permission_prompt | $HOME/.claude/scripts/claude-notify.sh "Claude" "권한 승인 필요" || true |
| Stop | * | bash $HOME/.claude/scripts/toolcall-text-guard.sh |
| Stop | * | PROJECT=$(basename "$PWD"); [ -z "$PROJECT" ] && PROJECT='Claude'; $HOME/.claude/scripts/claude-notify.sh "$PROJECT" "작업 완료" || true |
| SubagentStop | * | bash $HOME/.claude/scripts/toolcall-text-guard.sh |
| SubagentStop | * | bash $HOME/.claude/scripts/t41-subagent-stop.sh |
| UserPromptSubmit | * | bash $HOME/.claude/scripts/prompt-log.sh |
| SessionStart | * | bash $HOME/.claude/scripts/harness-usage-check.sh || true |
| SessionStart | * | OUT=$(bash $HOME/.claude/scripts/pattern-detect.sh 3 2>/dev/null); [ -n "$OUT" ] && printf '[T40 반복 패턴 후보 — 3회+ 검출, 스킬화/편입/룰화 제안 판단 대상 (이미 스킬로 커버되면 불요 판정)]\n%s\n' "$OUT"; true |
| SessionStart | * | bash $HOME/.claude/scripts/cost-clear-note.sh |
| SessionStart | * | bash $HOME/.claude/scripts/graphify-session-note.sh || true |
| SessionStart | * | bash $HOME/.claude/scripts/exchange-prune.sh |
| SessionStart | * | bash $HOME/.claude/scripts/correction-detect.sh || true |
| PreCompact | * | bash $HOME/.claude/scripts/cost-snapshot.sh |
| PreCompact | * | bash $HOME/.claude/scripts/read-once.sh --clear |
| PreCompact | * | bash $HOME/.claude/scripts/precompact-handover-check.sh |
| SessionEnd | clear | bash $HOME/.claude/scripts/cost-snapshot.sh |
Plugins (7)
superpowers@superpowers-marketplaceralph-loop@claude-plugins-officialcode-simplifier@claude-plugins-officialclaude-md-management@claude-plugins-officialclaudenews@claudenewscodex@openai-codexclaude-security@claude-plugins-official
Permissions
deny (18)
Bash(sudo:*)
Bash(shutdown:*)
Bash(reboot:*)
Read(./outputs/**)
Read(./lightning_logs/**)
Read(./wandb/**)
Read(./mlruns/**)
Read(./checkpoints/**)
Read(**/__pycache__/**)
Read(./node_modules/**)
Read(./dist/**)
Read(./build/**)
Read(**/*.ckpt)
Read(**/*.pth)
Read(**/*.safetensors)
Read(./_bench/**)
Read(./_cache/**)
Read(./_refs/**)
ask (0)
—
allow (10)
Read(**)
Glob(**)
Grep(**)
Bash(*)
WebSearch(*)
WebFetch(*)
Bash(tmux new-session *)
Bash(tmux send-keys *)
Bash(tmux capture-pane *)
Bash(bash ~/.claude/tools/train-watcher/launch-set.sh *)
Similar rigs
Scrollout/orbit
Orbit: a Claude Code plugin that gives one person a small software team. Role agents, a /decide PM, an operator running worker subagents in parallel, proof-checked TDD and a session ledger. A guard denies destructive git/GitHub actions, ask
Orchestrator 1.6k tok · · Q5
blueclover22/mw_agent_kit
Spec-driven development workflow for Claude Code — built around an approved design doc, where each stage waits for your approval.
Orchestrator 3.0k tok · · Q3
YounusHussain9/claude-agent-workspace
Multi-repo workspace for Claude Code: plain-language requests, specialist AI agents, approval gates and safety hooks.
Fort Knox 2.0k tok · · Q54
latuconsinafr/claude-code-config
The claude code config, tailored to myself, battle-tested (might inspires you)
Orchestrator 6.8k tok · · Q23