PM Agent Skill

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.

What This Is

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.

Phases

| # | 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 |

Install

Via npm

npm install -g @jahonn/agentskills-cli
agentskills install ./pm-agent-skill -t all

Via ClawHub (OpenClaw)

clawhub install ai-pm-agent

Manual

cp -r pm-agent-skill ~/.claude/skills/pm-agent

Usage

Full workflow

"I want to build an app that does X" → runs all 4 phases
"Run pm-agent on [problem statement]"

Single phase

"/research [product idea]"
"/define [research findings]"
"/validate [PRD]"
"/launch [validated product]"

Partial workflow

"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

Example

Input:

Small business owners struggle to track invoices across multiple payment platforms, leading to cash flow blind spots.

Output (4 phases, ~20 min):

1. DISCOVERY.md — JTBD analysis, 3 competitors mapped, TAM/SAM/SOM, 2 personas
2. PRD.md — Opportunity tree with 5 opportunities, RICE scored, press release + FAQ, 8 user stories
3. EXPERIMENT.md — 3 riskiest assumptions identified, Lean BML experiments with pass/fail criteria
4. GTM.md — ICP, positioning statement, 90-day OKRs, release checklist

Frameworks Used

| 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 |

Human-in-the-Loop

Each phase ends with a checkpoint:

• Approve — proceed as-is
• Edit — modify the artifact, then continue
• Rerun — provide feedback, regenerate

AI drafts. Humans decide.

Related

• Dev Workflow Skill — development sprint workflow
• Agent Skills Spec — the open standard
• Convoke Agents — 7-agent product discovery (inspiration)

License

MIT