Compound
Capture session learnings as structured insight markdown so future agent runs reuse anti-patterns, best practices, and replication steps from your knowledge base.
Overview
Compound is a journey-wide agent skill that turns session learnings into structured insight markdown with frontmatter, replication guides, and linked anti-pattern versus best-practice patterns—usable whenever a solo buil
Install
npx skills add https://github.com/ai-native-camp/camp-2 --skill compoundWhat is this skill?
- YAML frontmatter with domain, date, insight_type, component, context, key_learning, impact, and tags for machine-readabl
- Six-section insight body: Context, What We Tried, What Worked, Why This Works, Replication Guide, and Related
- Numbered pattern cards with parallel anti-pattern and best-practice blocks (situation, action, result) plus Why and 적용 시
- Replication Guide with 적용 조건, 단계별 절차, 주의사항, and 변형 for repeatable agent workflows
- Cross-links to knowledge/[category]/[filename].md so compound docs stay tied to canonical reference material
- Six named insight sections in the document template (Context through Related)
- Eight frontmatter fields including domain, insight_type, key_learning, and impact
- Parallel anti-pattern and best-practice blocks with three fields each (situation, action, result)
Adoption & trust: 1.1k installs on skills.sh; 17 GitHub stars; 3/3 security scanners passed (skills.sh audits).
What problem does it solve?
You finish a useful agent session but the lesson lives only in chat, so the next task repeats the same dead ends.
Who is it for?
Indie builders maintaining a personal or repo knowledge/ folder who want every meaningful experiment encoded for Claude Code, Cursor, or Codex to reuse.
Skip if: Teams that only need informal notes in issues or wikis with no schema, or installs where you will not maintain linked knowledge markdown paths.
When should I use this skill?
After completing meaningful work or experiments when you have a concrete key_learning worth storing, when defining or referencing numbered patterns in knowledge/, or when an agent must always consult documented anti-patt
What do I get? / Deliverables
You get a consistent insight file under your knowledge tree with measurable results, replication steps, and pattern links agents can reference on the next invoke.
- One insight markdown file with valid YAML frontmatter and filled narrative sections
- Optional pattern reference blocks with links to knowledge/[category]/[filename].md
- Replication Guide section with applicable conditions, steps, caveats, and variants
Recommended Skills
Journey fit
Useful at every journey phase - explore requirements and options before committing to a direction.
Where it fits
After a throwaway prototype proves a UX path, record what failed in What We Tried and the winning flow in Replication Guide before you scope the real build.
When an API integration needs a non-obvious retry pattern, document it as workflow_pattern with best-practice triplets linked from knowledge/integrations/.
Capture a code-review anti-pattern (situation, action, result) so the next PR pass invokes the same pattern card instead of re-litigating style debates.
Log which distribution experiment moved a metric in What Worked and tag impact high so later content agents pick the same channel playbook.
After an incident fix, file an insight with component and insight_type enums so monitoring-related invokes pull the replication steps immediately.
How it compares
Use instead of ad-hoc “lessons learned” paragraphs in SKILL.md with no frontmatter enums or replication checklist.
Common Questions / FAQ
Who is compound for?
Compound is for solo and indie builders who run AI coding agents daily and want durable, searchable insight documents—not ephemeral thread history—for work, learning, projects, tools, or personal experiments.
When should I use compound?
Use compound after you validate an approach during Build integrations, when Ship debugging reveals a repeatable fix, when Launch distribution tests teach a channel rule, when Grow analytics changes your workflow, or during Operate iteration when production behavior surprises you—
Is compound safe to install?
Compound is a documentation and schema skill that guides markdown authoring and filesystem paths; review the Security Audits panel on this Prism page before installing and avoid pointing agents at secrets in context or tags.
SKILL.md
READMESKILL.md - Compound
## N. [패턴 이름] (항상 참조 필수) ### 안티패턴 - 상황: [이런 상황에서] - 행동: [이렇게 하면] - 결과: [이런 나쁜 결과가 발생한다] ### 베스트 프랙티스 - 상황: [이런 상황에서] - 행동: [이렇게 해야 한다] - 결과: [이런 좋은 결과가 나온다] **Why:** [이 패턴이 중요한 구조적 이유] **적용 시점:** [언제 이 패턴을 참조해야 하는가] **근거 문서:** `knowledge/[category]/[filename].md` --- domain: [work|learning|project|tool|personal] date: [YYYY-MM-DD] insight_type: [enum from schema] component: [enum from schema] context: "[1-3문장: 어떤 상황에서 이 인사이트가 나왔는지]" key_learning: "[1문장: 핵심 교훈]" impact: [critical|high|medium|low] tags: [keyword1, keyword2, keyword3] --- # Insight: [명확한 인사이트 제목] ## Context [배경 설명: 무엇을 하고 있었고, 어떤 목표가 있었는가] - 시기: [언제] - 상황: [어떤 프로젝트/활동 중이었는가] - 목표: [달성하려던 것] ## What We Tried **접근 1:** [설명] - 결과: [어떤 결과가 나왔는가] - 판단: [왜 효과적이지 않았는가] [첫 시도에 성공한 경우:] **직접 적용:** 첫 시도에서 효과가 확인됨. ## What Worked [실제로 효과가 있었던 방법] **구체적 행동:** - [행동 1] - [행동 2] **측정 가능한 결과:** - [지표 또는 정성적 결과] ## Why This Works 1. 근본 패턴: [왜 효과적인가] 2. 구조적 이유: [작동하는 구조적 메커니즘] 3. 전제 조건: [이것이 작동하기 위한 조건] ## Replication Guide - 적용 조건: [언제 이 방법을 쓸 수 있는가] - 단계별 절차: [구체적으로 어떻게 하는가] - 주의사항: [빠지기 쉬운 함정] - 변형: [상황에 따른 변형 방법] ## Related - [관련 문서 링크] [없으면:] 관련 문서 없음. # Compound — Insight Documentation Schema required_fields: domain: type: enum values: [work, learning, project, tool, personal] description: "인사이트가 속하는 영역" date: type: string pattern: '^\d{4}-\d{2}-\d{2}$' description: "인사이트가 도출된 날짜 (YYYY-MM-DD)" insight_type: type: enum values: - workflow_pattern # 효율적인 작업 흐름 발견 - problem_solving # 문제 해결 과정에서의 교훈 - tool_discovery # 도구/기능 활용법 발견 - automation_pattern # 자동화/효율화 패턴 - communication_pattern # 소통/협업에서의 교훈 - decision_learning # 의사결정에서의 교훈 - debugging_insight # 디버깅/트러블슈팅 교훈 - design_pattern # 설계/구조 패턴 - learning_method # 효과적인 학습법 발견 - productivity_hack # 생산성 향상 방법 - mistake_lesson # 실수에서 배운 교훈 - best_practice # 검증된 모범 사례 description: "인사이트의 유형" component: type: enum values: # work - process - collaboration - meeting - documentation - planning # learning - skill-building - knowledge-management - study-method - mentoring # project - architecture - implementation - testing - deployment - code-review # tool - claude-code - mcp - ide - cli - automation # personal - habit - focus - time-management - reflection description: "영역 내 하위 컴포넌트" context: type: string min_length: 20 max_length: 300 description: "어떤 상황에서 이 인사이트가 나왔는지 (1-3문장)" key_learning: type: string min_length: 10 max_length: 200 description: "핵심 교훈 한 문장" impact: type: enum values: [critical, high, medium, low] description: "인사이트의 영향 범위와 중요도" tags: type: array[string] min_items: 1 max_items: 8 description: "검색용 키워드 (lowercase, hyphen-separated)" optional_fields: related_components: type: array[string] description: "'domain/component' 형식" evidence: type: array[string] status: type: enum values: [validated, hypothesis, superseded, archived] default: validated source_event: type: string action_items: type: array[string] domain_component_mapping: work: [process, collaboration, meeting, documentation, planning] learning: [skill-building, knowledge-management, study-method, mentoring] project: [architecture, implementation, testing, deployment, code-review] tool: [claude-code, mcp, ide, cli, automation] personal: [habit, focus, time-management, reflection] category_mapping: workflow_pattern: "knowledge/workflow-patterns/" problem_solving: "knowledge/problem-solving/" tool_discovery: "knowledge/tool-discoveries/" automation_pattern: "knowledge/tool-discoveries/" communication_pattern: "knowledge/communicati