hplan

2-step 디자인 파이프라인: --step brief (타겟 분석 → DESIGN_BRIEF.md 생성) → --step token (토큰/DESIGN.md 생성). --step all이 기본값.

Design an agent memory system — working memory, episodic memory, semantic memory, and procedural memory. Use when building agents that need to remember context across sessions, learn from interactions, or maintain persistent knowledge. Covers storage strategy, retrieval patterns, and context management.

Select and design the right orchestration pattern for multi-agent systems. Evaluate Sequential, Parallel, Router, and Hierarchical patterns against your use case requirements. The Router pattern covers both agent selection (classify input to the right specialist) and model selection (route tasks to the right LLM by complexity to balance cost, latency, quality, and fallback chains). Use when deciding how multiple agents should coordinate, share context, delegate tasks, or which model each task should run on.

전략 설계 통합 — 비즈니스 모델 캔버스(biz-model), 경쟁 해자 분석(moat), 성장 루프 설계(growth-loop) 통합. Use when defining business strategy, competitive positioning, or growth mechanics for an AI product.

에이전트 환경 설정 통합 — 7요소 인스트럭션 작성(agent-instructions)과 CLAUDE.md/AGENTS.md 구성(claude-md) 통합. 에이전트 정체성·도구·제약·실패 모드 정의부터 프로젝트 메모리 파일까지 한 스킬로. Use when setting up a new agent or updating agent instructions.

PM이 사람에게 질문하고 답을 모으는 비동기 채널 — comms MCP(Gmail/Notion/Zoom/Slack)를 감싸는 번역기. --mode ask(질문 초안 작성), --mode pull-answers(스레드·회의록에서 답 수집), --mode digest(수집한 답 요약 → decision-log/ticket-bridge 라우팅), --mode review(PRD 스테이크홀더 리뷰 — 리뷰어 배정·코멘트 수집·Signoff audit trail), --mode solo(팀원 없을 때 Claude가 역할 대리 시뮬레이션), --mode init(팀 세팅 대화형 온보딩). 메시지를 자동 발송하지 않는다 — 초안/코멘트까지만. 솔로 유저는 solo, 기업 PM은 review로 분기. Use when a PM needs to ask teammates for status/decisions, run a multi-stakeholder PRD review, and collect answers into hplan.

Run a full product build loop in one orchestrated session — discover → research → design → PRD → task decomposition → team-based implementation. Use when the user invokes /build, when an idea needs end-to-end execution from problem to shipped change, or when a feature crosses discover/architect/deliver boundaries that would otherwise require manual hand-offs.

태스크별 fresh subagent 디스패치 + 2단계 게이트(spec→quality) 반복 실행. harness-plan 승인 후 구현 루프를 돌릴 때 사용. 기존 parallel-team(현 conductor에 통합)의 역할 병렬 방식과 달리, conductor는 태스크 순차+게이트다.

Write a complete unified PRD covering user/JTBD/decisions/scope/agent-spec/metrics/hypotheses in 15 sections. Single source of truth for both customer-facing products and the LLM agents inside them. Replaces the older 7-section agent-only template. prd is the canonical owner of §6 Now/Next/Later. --mode design-shotgun reads §1+§11 from existing PRD and generates harness/design-variants/ (4 HTML variants + comparison.md). --mode roadmap is the §6 sub-mode: generate(Mermaid gantt + ROADMAP.md), rice(deterministic RICE scoring), prioritize(Now/Next/Later 재분류).

docs/PRD.md를 파싱해 harness/QA_CHECKLIST.md를 자동 생성. ICP/실패 시나리오 기반으로 TC를 critical/major/minor 3등급으로 분류하고 디바이스·환경 링크. deliver 완료 후 또는 harness-build --step quality-gate 전에 실행.

2-mode UI respect skill — brief (generate RESPECT.md design brief before coding) and checkpoint (final pre-ship gate enforcing user-respect via α/β/γ gate matrix). Use --mode brief when starting any UI screen; use --mode checkpoint before shipping. RESPECT.md absence blocks craft-lint (exit 2).

스프린트 계획-실행-추적 통합 — 딜리버리 플랜 작성(delivery-plan)과 진척 추적(track) 통합. PRD → WBS 분해, predicted.json 초기화, probe/detect/report/checkpoint 실행. Use when planning or tracking a delivery sprint.

GitHub Issues / Linear / Jira ⇄ hplan 실행 레이어(sprint/.track) 번역기. --mode pull(Issues→WBS 후보), --mode estimate(predicted.json p50/p90 → 이슈 코멘트), --mode status(.track+git/PR 상태 + CI/CD + PR review → 이슈 코멘트), --mode push(WBS 태스크 → 이슈 생성). --system github|linear|jira로 대상 시스템 선택. --batch 플래그: write-back 확인 게이트를 전체 요약 1회로 묶음 (개별 게이트 스킵). 추정치를 직접 계산하지 않고 sprint 산출물을 전달만 한다. Use when syncing GitHub/Linear/Jira tickets with hplan sprint tracking, or when a PM wants estimates/progress written back onto issues.

Unified UI validation skill — hierarchy (Playwright DOM saliency + WCAG AA), motion (CSS transition vs RESPECT.md drift), drift (pHash N-screen consistency), mobile (375/768/1440px breakpoint). Each check is independently runnable and independently failable. --check 인자 미명시 시 에러 출력 후 사용 가능한 check 목록 안내 — auto-run 절대 금지., tc-gate (QA_CHECKLIST.md TC-ID별 Playwright 스크린샷 증거 생성 → harness/ui-evidence/)

Identify and prioritize the riskiest assumptions in an agent idea across four axes: Value, Feasibility, Reliability, and Ethics. Use after defining an agent opportunity and before starting implementation. Prevents building agents that work technically but fail operationally or cause unintended harm. Includes build-or-buy vendor decision framework.

Simulate and forecast agent operating costs before building. Model token consumption, API call frequency, and monthly burn rate across different models and usage patterns. Use when evaluating agent feasibility, setting cost KPIs, or comparing build vs buy economics. Prevents the 'it's just API calls' cost surprise.

인터뷰 대상자 확보 + 컨택 초안 생성 + 인터뷰 질문 설계. --mode plan(확보 전략), --mode linkedin(LinkedIn cold DM 초안), --mode community(커뮤니티 포스팅 초안), --mode survey(설문 초안), --mode interview-questions(인터뷰 질문 세트 설계). harness/pain.md를 채우기 위한 선행 단계. Use when a PM needs to find and contact interview candidates before evidence-gate.

Design where and how humans should intervene in agent workflows. Define automation boundaries, escalation triggers, and approval gates. Use when building agents that make consequential decisions, handle sensitive data, or operate in domains where errors have high impact. Prevents the 'fully autonomous' default trap.

Analyze where AI agents can add value and which tasks to automate — systematically map repetitive workflows, manual processes, and operational bottlenecks to identify the best agent opportunities. Build an Agent Opportunity Tree from desired outcomes to solvable problems, agent solution candidates, and validation experiments. Use when exploring where AI agents could add value to a platform or service, finding automation opportunities in workflows, identifying repetitive tasks worth automating, prioritizing which agent to build first, or mapping the full opportunity space before committing to development. Applicable to any domain — customer support, edtech, SaaS, operations, and more.

어떤 결정·아이디어든 AI에게 시키기 전에, AI가 나를 먼저 심문하게 만드는 소크라테스 질문법 도구. 6 질문유형 + 소크라테스식 루프 + 3 심화질문 + CoT 판단구조로 '사고 검증 질문 세트' 1장을 만든다. Use when 정답이 하나가 아닌 판단(기획·가격·채용·투자·제품 범위 등)을 앞두고, 작은 PRD/기획서/리서치를 쓰기 전에 가정을 먼저 점검하고 싶을 때. 단순 사실 질문엔 쓰지 않음.

Phase 0 Worth-Building Check + Phase 1 대화형 설계 + Phase 2 Signal Gate Bootstrap. 아이디어를 validated 설계 문서로 전환. deliver/prd 스킬 진입 전 필수 단계. 3문 PROCEED/WARN 판정으로 만들 가치를 먼저 확인한다.

Executable COGS gate for AI products. Runs a deterministic Python sampler (lognormal token-cost distribution) to compute p50/p90 per-paid-user monthly COGS, gross margin scenarios, and free-user abuse blend. Returns GREEN / CONDITIONAL_GO / RED before any paid AI product is greenlit. Use when promising a paid AI feature, comparing providers (Anthropic/OpenAI/Google), or when discover/cost-sim has produced a usage hypothesis and you need real numbers.

Append-only build/interview/pivot/hold/CONDITIONAL_GO decision log with 3-6 month self-eval audit. Records every gate decision with score + reasons; later backfilled with outcome (shipped, killed, alive_no_revenue, pivoted, external_success) to compute hit_rate, false_holds, and missed_builds. The only PM gate skill that measures its own accuracy over time.

Score a product idea against the 100-point evidence rubric before any PRD work. Eight axes: ICP specificity, recent painful event, current workaround, repetition, economic pain, switching trigger, MVP narrowness, and acquisition path to first 5 users. Returns build/interview/pivot/hold decision plus the specific axes that are weak. Use when a founder or PM is excited about an idea but evidence is thin, or before approving any spec-driven coding workflow (Spec-Kit, Kiro, GStack, Superpowers).

Append-only Do-Not-Build registry. Each exclusion carries a reason, an owning competitor, and a reopen_trigger that says what evidence would unblock it. Future runs auto-check new ideas against the registry with Korean-aware char-bigram fuzzy match. Use when an idea overlaps with an established competitor (any established competitor in your space), when a previous pivot was killed, or when you want a project's institutional memory to survive across PMs.

Export an approved Build Gate brief to the downstream coding ecosystem you actually use — Spec-Kit (specs/NNN-slug/{spec,plan,tasks}.md), Kiro (.kiro/specs/<slug>/{requirements,design,tasks}.md), GStack (/office-hours brief), or Claude Code (AGENTS.md + CLAUDE.md). Use when Evidence + Product + Build Gate have all been approved and you're ready to start implementation in your coding agent of choice.

Import AI synthesis output (BuildBetter MCP, Perspective AI, and similar AI synthesis tools) into hplan, then force a human to tag each quote with strength (strong/medium/weak) and Push/Pull/Habit/Anxiety/workaround/trigger axes. AI extracts quotes; humans assign evidence strength. Audits the SKILL.md rule: 5 interviews with 3 distinct strong-Push signals → proceed to Product Gate.

Generate a Teresa Torres-style Opportunity Solution Tree as docs/OPPORTUNITY_TREE.md with auto-rendered Mermaid diagram. Forces the discipline: opportunities are unmet user needs (not solutions), each solution links to exactly one experiment with a decision rule, and opportunities backed by fewer than 3 strong-Push interviews are flagged or pruned. Use as the Product Gate's primary artifact, after interview-synthesis has tagged enough strong-push quotes.

Respond to and learn from AI agent incidents — triage severity, coordinate response, contain blast radius, and write postmortems. Use when an agent produces harmful outputs, costs spike unexpectedly, accuracy drops suddenly, or users report critical failures.

Design the metrics hierarchy and OKRs for an AI agent — North Star, KPI derivation, and OKR setting. Supports --step north-star | kpi | okr | all (default).

주간/월간 운영 리뷰 + 이해관계자 업데이트 보고. 비용 추적(burn-rate)·주간 롤업(weekly-rollup)·실제 LLM 비용 vs COGS 대조·이상 감지·예산 권고. 추가로 이해관계자 보고서 4종: exec-summary(임원 1-pager), weekly-update(팀 주간), partner-brief(외부 파트너), confluence-export(사내 위키 포맷 변환). 수치 집계=결정론, 산문 생성만 LLM. Use when running regular operational reviews or communicating project status to stakeholders.

Interface with the PM-ENGINE-MEMORY file — the operator's accumulated PM tacit knowledge database. Enables agents to reference, search, and apply TK (Tacit Knowledge) entries, and supports TK extraction from experience (--mode extract), TK querying and referencing (--mode query), and TK-to-instruction conversion (--mode build). The core of the pm-engine competitive moat. --mode decide for pattern-matching against stored PM decision patterns. --mode save-decision for PRD-linked tech decision logging (harness/tech-decisions/TD-NNN.yaml). --mode index-codebase for scanning project files and surfacing unrecorded decision candidates.

에이전트 포트폴리오 관리 — 단일 에이전트 추적(agent-portfolio)과 포트폴리오 전체 리포트(portfolio-report) 통합. 에이전트 상태 카드, 크로스-에이전트 비용 비교, 포트폴리오 헬스 스코어. Use when managing multiple deployed agents.

Systematically review and improve AI agent reliability — identify failure patterns, assess error handling, design safeguards, and set reliability targets. Use when agents are producing inconsistent results, after incidents, or when preparing for production deployment.