4-agent product management workflow covering the full PM lifecycle: Research → Define → Validate → Launch. Built on 7 proven frameworks including JTBD, Opportunity Solution Tree, Amazon Working Backwards, and Lean Startup.
Works with any Agent Skills compatible tool: Claude Code, OpenAI Codex, Trae, Cursor, Gemini CLI, and more.
Every PM tool does one thing well: ChatPRD writes PRDs, Dovetail does research, Linear tracks tasks. But nobody connects them. You still manually shuttle context between tools.
This skill gives you a complete PM workflow as AI subagents — each phase produces a structured artifact that feeds the next. One problem statement in, full product plan out.
| # | Phase | Agent Does | Framework | Output |
|---|---|---|---|---|
| 1 | Research | JTBD analysis, competitor map, market sizing, personas | JTBD + Design Thinking | DISCOVERY.md |
| 2 | Define | Opportunity tree, RICE scoring, Amazon PRD, user stories | Opp. Tree + Amazon Working Backwards | PRD.md |
| 3 | Validate | Assumption mapping, experiment design, prototype plan | Design Sprint + Lean BML | EXPERIMENT.md |
| 4 | Launch | GTM strategy, OKRs, release checklist, feedback loop | Dual-Track Agile + OKR | GTM.md |
npm install -g @jahonn/agentskills-cli
agentskills install ./pm-agent-skill -t allclawhub install ai-pm-agentcp -r pm-agent-skill ~/.claude/skills/pm-agent"I want to build an app that does X" → runs all 4 phases
"Run pm-agent on [problem statement]"
"/research [product idea]"
"/define [research findings]"
"/validate [PRD]"
"/launch [validated product]"
"Just do a competitor analysis" → Research only
"Help me prioritize my backlog" → Define (RICE section)
"Write a PRD for this feature" → Define with feature description
Input:
Small business owners struggle to track invoices across multiple payment platforms, leading to cash flow blind spots.
Output (4 phases, ~20 min):
DISCOVERY.md— JTBD analysis, 3 competitors mapped, TAM/SAM/SOM, 2 personasPRD.md— Opportunity tree with 5 opportunities, RICE scored, press release + FAQ, 8 user storiesEXPERIMENT.md— 3 riskiest assumptions identified, Lean BML experiments with pass/fail criteriaGTM.md— ICP, positioning statement, 90-day OKRs, release checklist
| Framework | Phase | What It Does |
|---|---|---|
| Jobs-to-be-Done | Research | Understand what "job" users hire the product for |
| Design Thinking | Research | Empathize → Define → Ideate framework |
| Opportunity Solution Tree | Define | Map outcomes → opportunities → solutions |
| RICE Scoring | Define | Prioritize by Reach × Impact × Confidence / Effort |
| Kano Model | Define | Classify features: must-have / performance / delighter |
| Amazon Working Backwards | Define | Write press release before PRD |
| Design Sprint | Validate | 5-day prototype testing methodology |
| Lean Startup BML | Validate | Build → Measure → Learn experiment cycles |
| Dual-Track Agile | Launch | Discovery + Delivery running in parallel |
| OKR | Launch | Objectives and Key Results for 90-day sprints |
Each phase ends with a checkpoint:
- Approve — proceed as-is
- Edit — modify the artifact, then continue
- Rerun — provide feedback, regenerate
AI drafts. Humans decide.
- Dev Workflow Skill — development sprint workflow
- Agent Skills Spec — the open standard
- Convoke Agents — 7-agent product discovery (inspiration)
MIT