pm-os

PM Operating System

My personal AI-native product management operating system, built as a Claude Code plugin. It encodes the PM frameworks I actually use into reusable slash commands and specialized agents — grounded in live product data from the repo's data/ directory, not templates or hypotheticals.

Why this exists

I manage multiple products as a solo founder. Repeatable PM work — PRDs, opportunity scoring, market scans, decision logging, weekly reviews — follows the same frameworks every time, but needs fresh product context each time. This plugin solves that: the frameworks are codified here, the context is read from markdown files in the product repo itself.

I also use this repo to develop the operating system itself — with a meta-agent for architectural guidance and an eval skill for quality auditing.

How I use it

I install this as a Claude Code plugin in each of my product repos. The skills and agents become available as slash commands in any repo, reading that product's data/ files automatically. One product per repo — the repo IS the product.

claude plugin marketplace add <path-or-github-url>

Skills

Skill	What it does
/fetch-context	Reads product context from data/ (decisions, personas, backlog, recent Signals, Market Landscape)
/write-prd	Writes a PRD using a 6-section framework, hydrated from data/
/evaluate-opportunity	Scores an opportunity on 4 dimensions — Pursue, Pursue narrow, Park, or Kill
/market-scan <market>	Scans competitive landscape with parallel web search and appends findings to data/knowledge/market-landscape/ and data/signals/active.md
/break-down	Decomposes a PRD into kanban-ready work items using JTBD framing
/weekly-review	Weekly review: what shipped, what's blocked, Action Required signals, what's next. Portfolio mode supports multi-repo rollup.
/log-decision	Writes a structured decision file to data/decisions/
/log-signal	Appends a time-stamped observation (user feedback, competitive move, market signal, technical constraint, internal learning) to data/signals/active.md
/knowledge	Manages stakeholder profiles, reference docs, research, and Market Landscape entries in data/knowledge/
/tasks	Surfaces active tasks from data/tasks/active.md (runs at session start)
/define-persona	Writes a persona file grounded in evidence to data/personas/
/design-experiment	Structures a falsifiable experiment with pre-committed thresholds
/pricing	Structures a pricing decision: value metric, anchor, tiers, WTP validation plan
/sunset-product	Guided kill/park/graduate workflow with retro + archival
/memory-review	Reviews and prunes data/ entries for staleness
/pm-digest	Daily PM + AI news digest (internal, not exported)

Agents

Agent	Role
startup-advisor	Pressure-tests GTM, moat, unit economics (YC + McKinsey lens)
product-sculptor	Sculpts MVPs to atomic core, defines JTBD backlogs
growth-engineer	Distribution-first growth strategy, pre-launch funnels, copy
systems-architect	System design, APIs, infrastructure, security, AI/LLM systems

Agents collaborate via a one-hop protocol — any agent can consult one other agent for cross-domain input, using a shared scratchpad for handoff.

Setup

One MCP server is required:

• Tavily — web search and content extraction (for /market-scan and /pm-digest)

Steps:

1. Copy .mcp.json.example to .mcp.json
2. Replace API key placeholders with your keys
3. Restart your Claude Code session

Data layout (in the consumer repo)

Product data lives in markdown under data/:

data/
├── signals/active.md              # H3 per signal, newest first, inline metadata
├── decisions/
│   ├── index.md                   # One-line-per-decision manifest
│   └── YYYY-MM-DD-slug.md         # One file per decision
├── knowledge/
│   ├── people/
│   ├── reference/
│   ├── research/
│   └── market-landscape/          # Living docs, `## Scan —` append-only
├── personas/
│   ├── index.md
│   └── {slug}.md                  # One file per persona
└── tasks/
    ├── active.md                  # Now / Next / Later checkboxes
    └── done.md                    # Completed tasks

See .claude/context/data-schemas.md for full frontmatter and file conventions.

Migrating from Notion

If you previously ran this plugin against Notion databases, run /migrate-from-notion once to export pages into the data/ layout. It's idempotent — safe to re-run.