> ## Content Index
> Fetch the complete content index at: https://claudefox.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Product Engineering: 45 free agent skills for shipping a product with Claude Code and Codex
- URL: https://claudefox.com/product-engineering-45-agent-skills-claude-code-codex/
- Published: 2026-11-15T16:24:30.000Z
- Updated: 2026-11-15T16:25:43.000Z
- Description: Product Engineering (phuryn/product-engineering) is a free, MIT-licensed collection of 45 agent skills in six plugin packs for Claude Code, Codex and other agents, covering what to build, building, independent review and release. 143 of its 144 commits are co-authored with Claude.
- Author: hupicake
- Tags: Built with Claude

Product Engineering is a free collection of 45 agent skills for people who build and ship a product with coding agents, from deciding what to build to taking payments in production. It is on GitHub at phuryn/product-engineering, MIT-licensed, by Pawel Huryn of The Product Compass, and installs as six plugin packs in Claude Code or Codex. On November 15, 2026 it had 604 stars and 20 releases in total, starting with v1.0.0 on November 2, and 143 of its 144 commits carry a Claude Co-Authored-By line.

The README contrasts it with collections written for engineers, which check the code line by line. This one is aimed at product managers, founders and designers who ship with agents, and at engineers who own a product rather than only its code. Its main idea is that the session that wrote the code never signs it off: a fresh reviewer, ideally from another model family, reads every change before it deploys.

## Quick facts

- What it is: 45 skills in six packs: product, ai-shipping, agent-teams, agent-harness, app-stack and work-humanizer.
- Works in: Claude Code and Codex as plugins; Cursor, Gemini CLI, OpenCode, GitHub Copilot and other agents that read SKILL.md folders through npx skills or by copying folders.
- Licence and cost: MIT, free.
- Latest release: v1.12.0, November 14, 2026.
- Requirements: Python 3.9 or newer for the skills that ship scripts, plus the CLI each one drives; npx skills needs Node.js 22.20 or later.
- Context cost: the 41 skills that start themselves add about 6,200 characters of descriptions, about 2,700 tokens in Claude Code, per the README. A skill's full text loads only when it runs.

## How to install

In Claude Code, the README's commands add the marketplace and then each pack you want:

```
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
```

These install for your user in every project. Add --scope project to keep them to one repository (written to .claude/settings.json, shared once committed) or --scope local for yourself only. Third-party marketplaces don't auto-update by default, so the README says to turn it on under /plugin, Marketplaces. In Codex, run codex plugin marketplace add phuryn/product-engineering, then codex plugin add product@product-engineering and the same for each other pack you want, then start a new session. For other agents, npx skills add phuryn/product-engineering picks skills interactively.

The README also offers a one-line route: ask your agent in chat to install the skills from phuryn/product-engineering for this project only, with auto updates.

To update by hand: claude plugin update <pack>@product-engineering in Claude Code, codex plugin marketplace upgrade product-engineering in Codex, or npx skills update. The README warns that a skill both copied into a folder and installed as a pack loads twice; keep one, preferably the pack.

## First setup

After installing, the README says to apply the agent-instructions skill once to wire your product's AGENTS.md or CLAUDE.md: your choices, a docs index and router rows that tell the agent which skill to apply for screens, a deploying push, a review loop or delegation. From then on, if you lose track, type /product:start-here (Codex: $product:start-here) and say what you have; it tells you where the work stands and what comes next.

Four skills start only when you type them: start-here, ai-shipping:ship-check, agent-harness:handover and agent-harness:retro. The other 41 start themselves when the work matches. The app-stack pack assumes a SaaS on Next.js, Supabase, Clerk and Netlify; the other packs are described by job: deciding, shipping, working with other agents, the agent setup, and writing.

![The collection's eight-step lifecycle chart: discover what to build, build and check, ship and learn, with the skills used at each step](https://claudefox.com/content/images/lab/2026-11-15/product-engineering-45-agent-skills-claude-code-co-img4-e2fa0659b3.png)

Image: phuryn/product-engineering (MIT)

## What stands out

- A reviewer that never saw the build. The independent-review skill picks Grok first, then a fresh Codex session, then a fresh Claude Code session (the order since v1.10.0), and repeats rounds until one finds nothing the change introduced above low severity and nothing high or critical of any age.
- Rules with receipts. AGENTS.md requires a rule that came from an incident to keep its story, dated, in the skill's receipts.md; the README counts about 700\. For example, the v1.12.0 notes say a product built with the skills went two weeks, 77 review rounds and nine development deploys without one browser test, so each stage now ends with committed end-to-end tests.
- Dated platform facts that get corrected. v1.10.0 reports that the Codex review command the skills had recommended still gave the reviewer 546 tools on Codex CLI 0.166.0, most from connected apps; the corrected command, which disables apps, plugins, web search and more by name, left 22.
- A budget for context. tools/validate.py, run in CI on pushes to main and next and on pull requests, holds the auto-loading descriptions to 6,200 characters in total and 300 each; at v1.11.0 they totalled 6,155.
- Opt-in for risky partners. The gemini-agents skill runs only after you agree, because Google's Antigravity CLI FAQ, read on November 13, names consumer plans as unsupported inside a third-party coding agent; the skill says the risk is yours.

## How it was built with Claude

Of the 144 commits on main on November 15, 143 carry a Claude Co-Authored-By line: 133 name Claude Opus 5.5 and 10 just Claude. All are authored by Pawel Huryn under two names. The skills themselves were learned on three live products by the author (AskOne, Staygreen and AFK Pilot), built with Claude Code, Codex and Grok; by the README's count 28 of the 45 skills come from that work and 14 are established methods such as the strategy frameworks.

The CLAUDE.md is a single line, @AGENTS.md, which imports AGENTS.md: about 2,600 words of rules for any agent maintaining the collection. Two stand out. Every skill must read as general guidance for strangers, so a product appears only by kind and a leak check scans every file for private names held in an uncommitted list. And lessons flow one way: a product records an incident in its own docs, the general lesson is ported into the collection, and products install the packs rather than keeping copies, after a four-day build produced about 140 lessons while its skill copies stayed byte-identical.

**Try it with your AI.** You can borrow the collection's review rule without installing anything. This prompt has your agent set up a fresh-reviewer step in your own repository.

```
I want my repository at [REPO PATH] (replace with the real folder) to follow one rule
from phuryn/product-engineering: the session that wrote a change never signs it off; a
fresh session that never saw the build reviews it before any deploying push.

1. First read my AGENTS.md or CLAUDE.md, if either exists, and tell me which coding
   agent CLIs are on my PATH (claude, codex, grok), using command -v only.
2. Propose a short section for my agent instructions that says: before a push that
   deploys, start a fresh reviewer session in read-only mode with a brief naming the
   change, the files and what to check; repeat rounds until a round finds nothing the
   change introduced above low severity and nothing high or critical.
3. Prefer a reviewer from a different model family than the builder; if only one is
   installed, use a fresh session of that one and tell me it shares more blind spots
   than a different model family would.
4. Show me the exact text and any command you would add, and wait for my confirmation
   before writing files.

Do not install plugins, change credentials or settings files, push, or delete anything.
```

## Known limits

- It is opinionated about stack. The app-stack pack assumes Next.js, Supabase, Clerk and Netlify, and the uptime-monitoring skill defaults to Staygreen, the author's own service; AGENTS.md says it is named with a disclosure each time.
- Grok is the default reviewer. Without the Grok CLI the review falls back to Codex, then Claude Code; v1.10.0 notes that a fresh session of the builder's own model shares more blind spots than another family would.
- Gemini is a terms risk. The skill says Google does not support it on consumer plans and that third-party access through an Antigravity sign-in can lead to suspension.
- Overlap with pm-skills. Five product skills also ship in the author's pm-skills collection, so each loads twice if both are installed and your agent picks either copy. Separately, the README says to turn off pm-ai-shipping once ai-shipping is installed, since that kit moved here.
- Project-scope installs across several git worktrees: the README warns an update may act on another worktree's record in \~/.claude/plugins/installed\_plugins.json (seen on Claude Code 2.1.319).
- A young, fast-moving project: 20 releases in 13 days, and all 144 commits by one author (Pawel Huryn, under two names). AGENTS.md says installs read main, so the method can change between sessions.

## Problems and feedback

Questions and bug reports go to the repository's GitHub issues (none open on November 15). Requests the collection decided not to take are listed with reasons in rejected.md; the README says what changed is in each commit, and each release has notes.

## FAQ

Is Product Engineering free? Yes. It is MIT-licensed, and the author's companion collection pm-skills is free too.

Does it work without Claude Code? Yes. Codex reads the same marketplace, and Cursor, Gemini CLI, OpenCode, Copilot and others can load the SKILL.md folders through npx skills or a manual copy.

How many tokens does it add? About 2,700 tokens in Claude Code for the 41 auto-starting skill descriptions, by the README's measure. Full skill text loads only when a skill runs.

What do I type first? /product:start-here in Claude Code or $product:start-here in Codex, then describe what you have: an idea, user evidence, a bug or a release.

Source: [GitHub: phuryn/product-engineering](https://github.com/phuryn/product-engineering?ref=claudefox.com), [Product Engineering releases](https://github.com/phuryn/product-engineering/releases?ref=claudefox.com), [Product Engineering AGENTS.md](https://github.com/phuryn/product-engineering/blob/main/AGENTS.md?ref=claudefox.com), [Product Engineering case studies](https://github.com/phuryn/product-engineering/blob/main/CASE-STUDIES.md?ref=claudefox.com), [Product Engineering README](https://github.com/phuryn/product-engineering/blob/main/README.md?ref=claudefox.com), [Product Engineering rejected.md](https://github.com/phuryn/product-engineering/blob/main/rejected.md?ref=claudefox.com)

This story first appeared in the [claudefox daily briefing for November 15, 2026](https://claudefox.com/daily-2026-11-15/).