artibot

Claude advisor tool strategy for cost-optimized quality retention. Sonnet executor + Opus advisor pattern reduces API cost by ~50% while maintaining near-Opus quality on complex judgment calls. Auto-activates when discussing model cost optimization, executor-advisor delegation, or advisor API integration. Triggers: advisor, cost optimization, model strategy, executor-advisor, advisor tool

Use when creating or editing a SKILL.md. 자연어 트리거: '스킬 만들어줘', '새 스킬 추가', '스킬 작성해줘', 'skill 만들기', '스킬 프론트매터 고쳐줘', 'create a skill', 'edit SKILL.md'.

로컬 관측성(observability) 스킬 — Claude Code hook 이벤트를 JSONL 큐로 fan-out하고 localhost WebSocket 브로드캐스트로 대시보드에 노출. 외부 전송 0건, Artibot 내부 전용. Auto-activates when: building a local hook monitor, an on-device agent dashboard, or an observability surface. Triggers: hook monitor, agent dashboard, event tap, local telemetry, claude code hooks, 훅 모니터, 이벤트 대시보드, 관측성.

Distill a person, role, or character into a reusable persona skill via the 6-layer schema. Use when the user wants to compile a colleague, an internal role archetype, or an authoring style into a persona-* skill — not when invoking an existing persona.

프롬프트 캐시 최적화 전략 — Dynamic Boundary 배치, 토큰 예산 관리, 정적·동적 영역 분리. 자연어 트리거: '프롬프트 캐시 최적화해줘', '토큰 예산 관리', '캐시 효율 높여줘', 'CLAUDE.md 레이아웃 정리'.

Dynamic tool approval predicates and the bash command allowlist. Use when the user asks about HITL approvals or tightening the pre-bash gate. Triggers: tool approval, approval predicate, bash allowlist, pre-bash, permission gate, HITL.

Coding standards and style guide enforcing immutability, error handling, file organization, and naming conventions. Auto-activates when: writing or modifying code, code review, creating new files or components. Triggers: code, write, edit, implement, component, function, class, style, 코드 작성, 코딩 스타일, 코드 리뷰, 구현, 컴포넌트, 함수

Auto-activates when the user asks to improve, enhance, audit, or propose changes without specifying a concrete problem — especially "개선", "보완", "발전방안", "제안", "감사", "전수조사", "업그레이드", "더 좋게", "추가하면 좋을", "enhance", "improve", "propose", "audit", "optimization", YAGNI, over-engineering, "필요한가", "어떻게 하면 더 좋아질까", "무엇을 더 보강할 수 있을까", "개선안 좀 찾아봐". Also fires when a candidate list of proposals exists and none has been cross-checked against live code evidence yet.

코드를 공격자 관점에서 리뷰하여 취약점, 엣지 케이스, 설계 결함을 탐지. Use when performing security review, attack surface analysis, pre-deploy verification, or PR security check, or mentions 적대적 리뷰, 공격적 리뷰, 취약점 리뷰.

Evidence-before-claims verification enforcing proof of completion before any "done" claim. Auto-activates when: completing tasks, reporting results, claiming success, writing summaries. Triggers: done, complete, finished, implemented, fixed, resolved, verified, shipped

Delegation strategies for parallel and complex multi-file operations using Sub-Agent or Team Mode. Sub-Agent Mode: Task tool for focused, single-session tasks. Team Mode: Agent Teams API for complex multi-domain coordination with peer communication. Auto-activates when: >7 directories, >50 files, multi-domain operations, high complexity tasks, team coordination needed. Triggers: delegate, parallel, sub-agent, team, concurrent, large-scale, orchestrate, coordinate, 위임, 병렬, 대규모, 팀

Context compaction timing strategy — suggests WHEN to trigger compaction based on context usage patterns. Auto-activates when: context >75%, PreCompact triggered, long sessions, multi-phase tasks. Triggers: strategic compact, context strategy, compress timing, long session, context window, memory pressure See also: compaction-survival (정보 보존 전략)

Federated Swarm Intelligence for collective learning across Artibot instances. Shares anonymized learning patterns (tool success rates, error signatures, team compositions) with the swarm network while preserving privacy through PII scrubbing and differential privacy. Enables all participants to benefit from collective experience without exposing individual data. Triggers: swarm, collective, federated, sync patterns, share learning, global weights, opt-in, contribution

Visual regression testing using SSIM-based screenshot comparison. Auto-activates when: visual testing, UI regression checks, screenshot comparison, component validation, CSS fix suggestions needed. Triggers: visual, screenshot, regression, baseline, pixel, SSIM, UI validation, component check

Discovery flowchart for Artibot's skill library. Use this as orientation when the user asks 'what skills do you have', 'how do I do X', 'what commands are available', or when picking the right tool for an unfamiliar task.

세션 상태를 압축 스냅샷으로 변환하여 서브에이전트 프롬프트에 주입. Auto-activates when: spawning sub-agents, delegating tasks, transferring session context between agents. Triggers: snapshot, session context, agent context, sub-agent, 스냅샷, 세션 컨텍스트, 서브에이전트, 컨텍스트 주입.

컴팩션 생존 전략 — 컨텍스트 압축 시 핵심 정보 보존 방법. Auto-activates when: context is about to compact, key files must survive compression, pending work needs to persist. Triggers: compaction, context loss, context compression, 컴팩션, 컨텍스트 손실, 컨텍스트 압축, 컨텍스트 윈도우. See also strategic-compact for compaction timing.

End-of-day retrospective dashboard of today's commits, quality, work status, and next steps. Use when the user wants a daily recap or retrospective. Triggers: daily, recap, 회고, 일일 보고, 오늘 작업, 오늘 뭐 했지, retrospective, daily report, 오늘 뭐 했는지 정리해줘, 일일 리뷰 해줘, 오늘 작업 요약해줘

Input/output and per-tool guardrails with tripwire semantics. Auto-activates when: an agent processes untrusted input, calls a sensitive tool, or must short-circuit on a policy violation. Triggers: guardrail, input validation, tripwire, tool guardrail, policy check, refusal, agent safety.

Provides shadcn/ui component patterns for React including installation, theming, dark mode, CVA variants, and accessibility with Radix UI primitives. Use when the user asks about building UIs with shadcn/ui, setting up component libraries, customizing themes or dark mode, or composing form patterns. Triggers: shadcn, shadcn/ui, radix, tailwind components, ui library, component library, dark mode.

Catalog of agent orchestration patterns: agents-as-tools, handoff with history filter, deterministic workflows, parallelization, and LLM-as-judge loops. Triggers: orchestration, multi-agent, handoff, agent as tool, agent topology, parallelize agents, LLM-as-judge.

Infrastructure automation and reliability engineering decision framework for deployment pipelines and observability. Use when user mentions deployment, CI/CD, Docker, Kubernetes, monitoring, infrastructure as code, or pipeline automation. Triggers: deploy, infrastructure, CI/CD, DevOps, pipeline, monitoring, 배포, 인프라.

Provides authentication and authorization patterns including JWT, OAuth2, RBAC, session management, and managed auth services like Auth0 and Clerk. Use when the user asks about login flows, access control, OAuth or OIDC, JWT tokens, or CSRF protection. Triggers: authentication, authorization, OAuth, JWT, RBAC, session, Auth0, Clerk.

ATLAS Quality Framework for Artibot: Automated, Tested, Learned, Adaptive, Secure. Defines the validation cycle, coverage targets, and GRPO-driven continuous quality improvement. Auto-activates when: reviewing code, setting quality gates, running a validation cycle, or assessing coverage. Triggers: quality framework, ATLAS, validation cycle, quality gate, coverage threshold, 품질 게이트, 커버리지, 품질 검토.

Executive summaries and data narratives with templates for weekly, monthly, and quarterly cadences. Use when user asks about executive summary, performance recap, weekly digest, monthly rollup, marketing report, 리포트, 성과 보고서, 주간 보고서, 월간 보고서, 성과 리포트 만들어줘, 이번 달 마케팅 보고서 써줘, or 임원 요약 만들어줘.

SEO strategy covering keyword research, search intent, ranking factors, keyword clustering, and GEO (Generative Engine Optimization). Use when user asks about SEO strategy, keyword research, search intent, ranking factors, content gaps, SEO roadmap, GEO, 검색엔진최적화, 키워드 리서치, 검색 의도, 키워드 뭐로 잡을지 알려줘, SEO 로드맵 짜줘, or 검색 상위 노출 전략 세워줘.

SPEC format for structured requirements, acceptance criteria, and technical specifications. Use when the user asks about requirements, acceptance criteria, or technical specs. Triggers: spec, requirements, specification, acceptance criteria, EARS, user story.

Test-first discipline for new features, bug fixes, and coverage work. Auto-activates when tests are requested, TDD is mentioned, or refactoring needs a safety net. Triggers: test, TDD, test-driven, coverage, unit test, red green refactor, 테스트, 커버리지, 테스트 먼저 써줘, TDD로 짜줘, 테스트 커버리지 올려줘

Parallel team execution with cross-check — the leader delegates, teammates work independently, then verify each other. Use when the user asks for parallel independent work with cross-verification. Triggers: team, parallel team, cross-check, verification, 팀, 병렬 팀, 팀원들, 병렬로.

GitHub Actions 심화 패턴 — matrix strategy, reusable workflows, composite actions, 시크릿 관리, 캐싱 전략. Use when designing CI/CD pipelines, GitHub Actions workflows, or deployment automation. 자연어 트리거: 'CI 파이프라인 만들어줘', '깃헙 액션 워크플로 짜줘', '배포 자동화 해줘', 'GitHub Actions 설정'.

Turns ambiguous user requests into precise, actionable specifications using hypothesis-based MCQ clarification. Auto-activates when ambiguity score exceeds threshold, multi-intent conflicts are detected, or the user explicitly requests clarification on scope, target, or constraints. Triggers: clarify, unclear, what do you mean, ambiguous, vague, specify, 명확히, 구체적으로, 뭘 원하시는지, 좀 더 자세히 말해줘, 어떤 걸 원하는 건지 모르겠어, 범위가 너무 넓어

Audit CLAUDE.md files in a repository against a 6-criteria quality rubric, output a scored report and diff preview, and require user approval before any Edit. 자연어 트리거: 'CLAUDE.md 점검해줘', '메모리 품질 감사', '프로젝트 메모리 검토해줘', 'claude.md audit'.

Provides a structured DESIGN.md schema and curated brand references for AI-native visual design specification. 자연어 트리거: '디자인 시스템 만들어줘', 'DESIGN.md 작성', '디자인 토큰 정의해줘', '비주얼 디자인 가이드', '컬러 팔레트 정해줘'.

Provides Mermaid diagramming patterns for technical documentation including flowcharts, sequence diagrams, ERDs, class diagrams, state diagrams, Gantt charts, and mindmaps. Use when creating architecture diagrams or documenting API sequences. 자연어 트리거: '다이어그램 그려줘', '머메이드로 그려줘', '시퀀스 다이어그램 만들어', '플로우차트 그려줘'.

k6 기반 성능·부하 테스트 가이드 — 시나리오 설계, 임계값 설정, CI/CD 통합. Use when designing load tests, stress tests, or performance benchmarks. 자연어 트리거: '부하 테스트 해줘', '성능 테스트 만들어줘', 'k6 스크립트 작성', '스트레스 테스트'.

통합 관측성 가이드 — Prometheus, Grafana, 분산 트레이싱, SLI/SLO, 알림 규칙. Use when setting up monitoring, tracing, or alerting. 자연어 트리거: '모니터링 설정해줘', '관측성 구성', 'SLO 정의해줘', '알림 규칙 만들어'.

Provides cloud database patterns for serverless PostgreSQL, real-time databases, and edge-compatible data access (Neon, Supabase, Firebase, PlanetScale). 자연어 트리거: '클라우드 DB 골라줘', 'Supabase 연동해줘', '서버리스 데이터베이스 설정', '커넥션 풀링 구성'.

Provides deployment patterns for modern applications: cloud platforms, Docker, Kubernetes, and CI/CD. 자연어 트리거: '배포 파이프라인 설정해줘', 'Vercel 배포해줘', 'Dockerfile 작성해줘', '쿠버네티스 배포 구성'.

Automatic session work journal maintained in auto-memory, recording tasks, decisions, and pending items for recovery. 자연어 트리거: '작업 기록해줘', '세션 로그 남겨줘', '오늘 한 일 정리', '결정 사항 기록'.

Quality enforcement for casual or natural-language coding requests. Use when handling natural language coding requests, multi-part instructions, or casual Korean/English coding commands like '해줘', '만들어', '수정해', '바꿔', '고쳐'.