pmstack

A PM toolkit that runs inside Claude. Turn customer quotes into PRDs, compare competitors, design eval suites, draft exec briefs — all from one consistent set of commands you type into Claude (in your terminal, browser, or mobile app).

Who this is for

You're a technical PM — comfortable in a terminal, can read code, ship product specs and metrics dashboards. You use Claude (or want to) but you're not deep in LLM-land. You don't want to memorize prompt engineering tricks; you want a co-pilot that already knows how a senior PM thinks.

If that's you, this saves you 5–15 minutes per artifact and stops you from staring at a blank page. If you're a senior PM who already has your own templates, fork this and make them yours.

What you get

Nine commands, each producing a real artifact (markdown or YAML in outputs/) you can paste into Notion or hand to a teammate.

You type…	You get…
/prd "Customers say onboarding takes 3 days"	A PRD draft from a single customer signal
/competitive AI coding assistants	3–5 competitors compared, with white-space analysis
/compare Cursor Windsurf	Feature-by-feature comparison with a built-in test plan
/metrics "search relevance for our agent"	A measurement framework: North Star, supporting metrics, counter-metrics
/brief "Q2 launch slipping" exec	A one-page stakeholder brief sized for the audience
/eval "support chatbot"	A test suite design for an AI feature (a "what would good look like")
/run-eval outputs/eval-...yaml	Actually runs the test suite and scores it
/sprint "Customers want pricing transparency"	All four — PRD → metrics → eval → brief — with you approving each step
/eval-self	Scores pmstack itself against its own quality bar

Don't worry about understanding all of them yet. Start with /prd on a real customer quote you have lying around — that's the fastest way to feel the value.

Show me what the output actually looks like

Browse the examples gallery → for real artifacts each command produces. The headliner:

• outputs/eval-claude-ultraplan-2026-04-24.yaml — what /eval produced when asked to design a test suite for a hypothetical "Claude Ultraplan" planning agent. 10 capabilities, 12 failure modes, 13 metrics, 15 test cases. This is the kind of artifact most teams ship AI features without — pmstack makes it 60 seconds of work.
• outputs/pmstack-roadmap-2026-04-24.md — strategic roadmap memo. Not from a skill directly, but written by the same Claude+pmstack pairing.
• outputs/verification-2026-04-24.md — a verification report. Format you can reuse for any "does this actually work?" review.

Install (one line)

curl -fsSL https://raw.githubusercontent.com/RyanAlberts/pmstack/main/install.sh | bash

That installs into the folder you're currently in. Want to install it elsewhere or make it available everywhere?

# Install into a specific folder
curl -fsSL https://raw.githubusercontent.com/RyanAlberts/pmstack/main/install.sh | bash -s -- ~/work/my-pm-stuff

# Install for every Claude Code session (global)
curl -fsSL https://raw.githubusercontent.com/RyanAlberts/pmstack/main/install.sh | bash -s -- --global

The installer downloads pmstack to a temp folder, copies the bits you need into your target, and cleans up after itself. Nothing leftover in your home directory. Read the script first if you'd like: install.sh.

Manual install (if you prefer to clone and inspect)

git clone https://github.com/RyanAlberts/pmstack.git
cd pmstack
./setup ~/work/my-pm-stuff

Get started in 60 seconds

Step 1. Open Claude Code in the folder you installed into.

cd ~/work/my-pm-stuff   # or wherever you installed
claude                    # starts Claude Code in this folder

Don't have Claude Code yet? It's Anthropic's terminal app for Claude — install with one command from claude.com/code. The pmstack commands only work inside Claude Code (or, for browser/mobile, see "Using this on your phone" below).

Step 2. Type one of the commands. Real example to try right now:

/prd "Three of our biggest customers said onboarding feels like a full-day project. Two churned in week one."

Claude will produce a full PRD draft and save it to outputs/prd-onboarding-friction-2026-04-25.md. Open it. Edit it. Hand it to engineering.

Step 3. That's it. The other commands work the same way. Each writes its output to outputs/ so you can find it later, share it, or feed it into the next command.

---

Using pmstack outside Claude Code

Slash commands only work in Claude Code (terminal). Most other AI tools work fine — the underlying skills are just markdown. Pick your tool and follow the 2-5 minute setup:

→ docs/using-other-tools.md covers Claude.ai (web/mobile/desktop), Cursor agent mode, ChatGPT, Gemini, and "anything else."

---

What's an eval? (the one piece of jargon you'll see a lot)

An eval is a test suite for an AI feature. You define inputs ("what a real user might say") and what good looks like, then run them through your AI system and score the results. It's how you know whether gpt-4 does this task better than claude-sonnet, or whether your latest prompt change made things better or worse.

Two pmstack commands handle this:

• /eval — Designs the test suite. Output: a YAML file describing what to test and how to grade it.
• /run-eval — Actually runs the test suite against a real AI system and writes a scored report.

Most PMs ship AI features without evals because designing one feels intimidating. /eval makes the design 80% done in 60 seconds. /run-eval will hard-stop if you haven't told it what AI system to test against — it never invents fake scores.

A walk-through for your first eval is in docs/run-eval-setup.md.

A few common questions

Do I need to know how Claude works internally? No. You should know what an LLM is and roughly that they have limits, but pmstack hides everything else.

Does this cost money? The commands themselves are free. The Claude tokens used to run them count against your Claude subscription (Pro/Max/Team) or API key. A typical PRD or competitive analysis is well under a penny. The /run-eval command can use more tokens — it tells you the estimate before each run and asks before spending.

What if I want to change how a command behaves? Open the file in skills/<name>.md (e.g., skills/prd-from-signal.md) and edit. The skill is just markdown — Claude reads it as instructions. Save and re-run; your changes apply immediately.

Is my data sent anywhere unusual? No. Everything runs through your existing Claude account. pmstack adds no servers, no telemetry, no third parties. The skills are just markdown files on your disk.

Want to contribute, fork, or learn how it's built?

• DECISIONS.md — every design choice and why
• outputs/pmstack-roadmap-2026-04-24.md — what's coming next
• Inspired by Gstack (the engineer's version) and Karpathy's LLM Wiki

---

Built by Ryan Alberts — Staff PM in Agentic AI. PRs and forks welcome.