Product Management Skill

AI-native product management for startups. Transform Claude into an expert PM that processes signals, not just feature lists.

Core Philosophy

WINNING = Pain × Timing × Execution Capability

Filter aggressively from 50 gaps to 3-5 high-conviction priorities. Expert PMs track signals with confidence scores, timestamps, and velocity.

Commands Quick Reference

Command Purpose

/pm:analyze

Scan codebase + interview for product inventory

/pm:landscape

Research competitor landscape

/pm:gaps

Run gap analysis with WINNING filter

/pm:file

Batch create GitHub Issues for approved gaps

/pm:prd

Generate PRD and create GitHub Issue

/pm:sync

Sync local cache with GitHub Issues

Agents

This plugin provides specialized agents for autonomous tasks:

Agent Triggers On Purpose

research-agent

"research [competitor]", "scout [name]" Deep autonomous web research

gap-analyst

"find gaps", "what should we build" Systematic gap identification with scoring

prd-generator

"create PRD for [feature]" Generate PRD + create GitHub Issue

Data Storage

All data stored in .pm/ folder at project root:

.pm/ ├── config.md # Positioning, scoring weights ├── product/ # Product inventory, architecture ├── competitors/ # Competitor profiles ├── gaps/ # Gap analyses with scores ├── requests/ # Synced GitHub Issues (for dedup) ├── prds/ # Generated PRDs └── cache/last-updated.json # Staleness tracking

See references/data-structure.md for complete file templates.

WINNING Filter Scoring

Hybrid scoring approach - Claude suggests researchable criteria, user scores domain-specific:

Criterion Scorer Source

Pain Intensity (1-10) Claude Review sentiment, support data

Market Timing (1-10) Claude Search trends, competitor velocity

Execution Capability (1-10) User Architecture fit, team skills

Strategic Fit (1-10) User Positioning alignment

Revenue Potential (1-10) User Conversion/retention impact

Competitive Moat (1-10) User Defensibility once built

Total: X/60 → Recommendation:

• 40+ → FILE (high conviction)

• 25-39 → WAIT (monitor)

• <25 → SKIP (not worth it)

See references/winning-filter.md for detailed scoring criteria.

Deduplication & Sync

Prevent duplicate feature requests by syncing with GitHub Issues:

On Session Start

• Check .pm/cache/last-updated.json for staleness

• If >24 hours since last sync, prompt for /pm:sync

/pm:sync Process

• Fetch all GitHub Issues with pm:* labels via gh issue list --json

• Update .pm/requests/[issue-number].md for each issue

• Update last-updated.json timestamp

Deduplication During Gap Analysis

• Load existing issues from .pm/requests/

• For each new gap, fuzzy match against existing:

• Title similarity (Levenshtein): 40% weight

• Keyword overlap: 30% weight

• Label match: 20% weight

• Description similarity: 10% weight

• Mark gaps as:

• EXISTING (>80% match) → Show linked issue

• SIMILAR (50-80%) → Warn, ask user

• NEW (<50%) → Proceed normally

Output Format

Flow Summaries

Product Analysis (/pm:analyze )

• Scan codebase for features (routes, components, APIs, models)

• Interview user for business context

• Generate inventory with technical moats and debt flags

• Save to .pm/product/

Competitive Intelligence (/pm:landscape , /pm:scout [name] )

• Research competitors via WebFetch/WebSearch

• Categorize features: Tablestakes, Differentiators, Emerging, Deprecated

• Save profiles to .pm/competitors/

Gap Analysis (/pm:gaps )

• Load product inventory + competitor profiles

• Check staleness (>30 days → prompt refresh)

• Sync with GitHub Issues for deduplication

• Identify all gaps, score with WINNING filter

• Mark as NEW/EXISTING/SIMILAR

• Save to .pm/gaps/[date]-analysis.md

Feature Filing (/pm:review , /pm:file )

• /pm:review

• Walk through gaps, decide FILE/WAIT/SKIP

• /pm:file

• Create GitHub Issues for approved gaps (skips duplicates)

• Apply labels: pm:feature-request , winning:* , priority:*

PRD Generation (/pm:prd [feature] )

• Load feature context from gap analysis or GitHub Issue

• Generate PRD: Problem, User Stories, Acceptance Criteria, etc.

• Save to .pm/prds/[slug].md

• Create GitHub Issue with PRD content as feature request

Backlog & Roadmap (/pm:backlog , /pm:roadmap )

• Fetch open issues with pm: labels

• Sort by WINNING score or RICE

• Organize into Now/Next/Later priorities

GitHub Integration

Labels (Auto-Created)

pm:feature-request pm:gap-identified pm:competitor-intel priority:now priority:next priority:later winning:high (40+) winning:medium (25-39) winning:low (<25)

Prerequisites

• GitHub CLI (gh ) installed and authenticated

• Run gh auth status to verify

See references/github-labels.md for label definitions. See references/issue-template.md for issue format.

Integration with spec-kit

This plugin handles WHAT to build and WHY (product discovery). For HOW to build it, use spec-kit:

PM Plugin → GitHub Issue → spec-kit /pm:file Creates issue /speckit.specify /pm:prd Creates issue /speckit.plan → /speckit.implement

The GitHub Issue IS the handoff—no separate command needed.

Staleness Handling

PM data ages. Handle proactively:

• Competitor data >30 days: Prompt refresh before gap analysis

• Gap analysis >14 days: Warn when viewing backlog

• GitHub sync >24 hours: Suggest /pm:sync on session start

⚠️ Competitor data is 45 days old. Run /pm:landscape to refresh.

Additional Resources

Reference Files

• references/winning-filter.md

• Detailed WINNING scoring criteria

• references/github-labels.md

• Label definitions and colors

• references/issue-template.md

• GitHub Issue template

• references/data-structure.md

• Complete .pm/ folder structure

Example Files

• examples/gap-analysis.md

• Sample gap analysis output

• examples/competitor-profile.md

• Sample competitor profile