ClaudeMap

Product Engineering

44 agent skills distilled from shipping three live products: decide what to build, build it with Claude Code or Codex, have a different model review it, release it — six plugin packs covering proposals, independent review, a release gate and SaaS production ops, for Claude Code, Codex, Cursor, Gemini CLI, OpenCode and GitHub Copilot (MIT)

Skills & Commandsskillspluginproduct managementcode reviewrelease gatesaas

Product Engineering is phuryn's (The Product Compass) open-source collection of agent skills (526 stars as of writing, 2026-11-12): of its 44 skills, 28 were learned while the author shipped three real products with Claude Code, Codex and Grok (AskOne, live Q&A and polls; Staygreen, answer monitoring; AFK Pilot), 14 are established methods, and the rest carry their own source labels. It is built for shippers who don't read code line by line: one artifact holds the research, the key screens and the open decisions, and the build reads what you accepted; the session that wrote the code never reviews itself — before every deploy-triggering push and every release, a fresh reviewer that has never seen the build (Grok by default, i.e. a different model family, or one you name) re-reviews it round after round until nothing newly introduced sits above low severity. The descriptions of the 40 self-starting skills stay loaded at about 6,000 characters (about 2,600 tokens in Claude Code); each skill's full text loads only when it runs, and platform facts carry dates. It works in Claude Code, Codex, Cursor, Gemini CLI, OpenCode, GitHub Copilot and any other agent that reads SKILL.md folders.

What it does

  • Six plugin packs, 44 skills: product (propose, competitor-analysis, market-segments and five more for deciding what to build), ai-shipping (ship-check, independent-review, release-gate and eight more), agent-teams (4), agent-harness (6), app-stack (14 SaaS production skills) and work-humanizer
  • A fresh reviewer that never saw the build re-reviews every deploy-triggering push and every release, round after round — Grok (a different model family) by default, another reviewer by name; no session reviews its own work
  • One artifact, not tickets: research, high-fidelity key screens and open decisions in one place; the build reads what you accepted, and what waits on you sits on one page, each item with a recommendation
  • Production is part of the method: legal-pages, saas-authorization, supabase-production, clerk-production, transactional-email, product-analytics and uptime-monitoring ship in the app-stack pack, not as an afterthought
  • Multi-agent roles built in: delegation defines the brief and what never delegates; codex-agents, claude-agents and grok-agents cast the others as peer, builder or reviewer
  • Self-validating: a validator checks every skill and every file a skill names; practice rules keep receipts (about 700 dated failure/measurement records), and a new rule needs an observed failure

When to use it

For anyone shipping a real product with coding agents: product managers, founders and designers who don't read code line by line, and multi-agent workflows where one session coordinates, another builds and a third reviews; also engineering teams who want release-gate and independent-review discipline without building the process themselves.

Getting started

# Claude Code (installs for your user, in every project):
claude plugin marketplace add phuryn/product-engineering
claude plugin install product@product-engineering
claude plugin install ai-shipping@product-engineering
claude plugin install agent-teams@product-engineering
claude plugin install agent-harness@product-engineering
claude plugin install app-stack@product-engineering
claude plugin install work-humanizer@product-engineering
# add --scope project to any line for this project only (.claude/settings.json), or --scope local for yourself here
# third-party marketplaces don't auto-update by default: /plugin → Marketplaces → product-engineering → enable auto-update
# then run /product:start-here and say what you have (in a terminal, if it isn't found: /reload-plugins)

# Codex:
codex plugin marketplace add phuryn/product-engineering
codex plugin add product@product-engineering   # and the other packs
# update: codex plugin marketplace upgrade product-engineering, then a new session

# any other agent (Cursor, Gemini CLI, OpenCode, Copilot, Windsurf ...):
npx skills add phuryn/product-engineering          # pick skills and agents interactively
npx skills update                             # later, to update
# npx skills needs Node.js 22.20+; skill scripts need Python 3.9+ and the CLI they drive

Things to know

Known pitfall, dated in the README (2026-11-07, Claude Code 2.1.319): with several worktrees of one repository, a plugin update can change another worktree's record in ~/.claude/plugins/installed_plugins.json and then report 'already at the latest version' while your worktree stays on the old commit — check each worktree's projectPath and gitCommitSha after updating. Five product skills (competitor-analysis, market-segments, value-proposition, beachhead-segment, growth-loops) exist in both this collection and the author's pm-skills and load twice if both are installed — your agent picks either copy; the AI Shipping Kit moved here from pm-skills, so disable pm-ai-shipping once these packs are in. A skill both copied into a skills folder and installed as a pack also loads twice — keep one, and prefer the pack, which updates. Roles install separately: the reviewer starts without build context on purpose, and a builder needs app-stack for screens and an MCP server. Per the README's own source labels, work-humanizer rests on a detector test and product-analytics on its vendors' docs. MIT, from The Product Compass (productcompass.pm). 526 stars as of writing (2026-11-12).

Repository

phuryn/product-engineering

Charted

Related on the map

  • An instruction-only skill that makes agents research before they ask, plan before they build, and verify before they call anything done.

    • workflow
    • verification
    • quality
    • skill
  • Guizang's agent skill turns your repo's recent updates into a finished promo video — real components, code-rendered animation, an original code-written soundtrack — no video editor involved.

    • agent-skill
    • video
    • promo
    • marketing
  • An agent skill that geolocates a photo with nothing readable in it — OSM geometry, elevation skylines and satellite tiles turn “where was this taken?” into a measured answer with an error radius.

    • agent-skill
    • geolocation
    • osint
    • openstreetmap