1117 professional Agent Skills + workflow recipes — searchable & fetchable over MCP.
PM Skills — 515 professional Agent Skills + workflow recipes — searchable & fetchable over MCP. This server exposes a catalog of skills and workflows for AI agents, enabling quick discovery and retrieval through the MCP protocol. It draws on a broad set of topics including product management, productivity, and prompt engineering.
🛠️ Key Features
515 professional agent skills and workflow recipes
Searchable and fetchable over MCP
Topics span product management, AI agents, and prompt engineering
Readme excerpt provided for quick context
🚀 Use Cases
Discover skills for Claude, ChatGPT, Gemini, Cursor, Codex, and Hermes
Integrate agent skills into MCP-powered pipelines
Explore prompts and workflows for AI agents in product teams
⚡ Developer Benefits
Grounded in MCP-server ecosystem
Clear taxonomy via topics for discovery
Readme excerpt offers quick project context
Suitable for plugin-directory style cataloging
⚠️ Limitations
ReadmeExcerpt appears truncated in source data; rely on repository for full content
Metadata fields limited to slug, name, description, topics, toolCount, and readmeExcerpt
🧠 PM Skills — 1153 Professional Agent Skills for Claude, ChatGPT, Gemini, Cursor, Codex & Hermes
Your landlord kept your deposit. Your mom got a medical bill that makes no sense. You got laid off on a Tuesday. Someone you love died, and no one handed you the checklist.
Generic AI gives you filler for the moments that matter most. PM Skills gives your AI the exact framework a senior professional would use — for 1,117 real tasks, across work and life.
👉 Start with your moment, not the catalogue → Skill Packs
🍼 New parent · 💼 Just laid off · 🌍 New to this country · 👵 Caring for a parent · 🕊️ Losing someone · 💸 Money in crisis · 🔑 Starting over · 🤖 Getting serious about AI
PM Skills is an open-source library of 1153 Agent Skills — plain-markdown SKILL.md files that teach an AI assistant to do one professional task to a senior professional's standard, from writing a PRD to decoding a lease or running a blameless postmortem. Each skill bundles the framework, an output template, quality checks, and anti-patterns. It is MIT-licensed and works with Claude, ChatGPT, Gemini, Cursor, and Codex.
Decode a lease before you sign it. Write a PRD your team can execute. Simulate the promotion committee before the real one meets. Check the weather with zero API keys. Generic AI gives you filler; these give you the structure a senior professional actually uses.
Works natively in Claude Code and Hermes Agent, with ready-to-paste exports for ChatGPT, Gemini, Cursor, Codex and 8 more tools. (PM stands for Professional, not just Product Management.)
🐣 New here? Pick a door — each takes about 30 seconds
Just looking → open the ▶ Playground and run a skill in your browser. Nothing to install, nothing to sign up for.
You use Claude Code → type /plugin, search pm-skills, install. Done — ask "decode this lease" and watch.
You use anything else →npx pm-claude-skills add and pick your tool from the menu (Cursor, Codex, Windsurf, ChatGPT, Gemini…).
Want the guided tour → browse the searchable catalogue site and subscribe to get an email whenever new skills launch.
Nothing here can scare your setup. A skill is a markdown file your AI reads — no runtime, no telemetry, no accounts. Installing copies text files; uninstalling is deleting them. Skeptical? Good instinct: read one first — it's designed to be read by humans too.
Don't know what to look for? Describe your task in plain words at 🔎 find — "my landlord kept my deposit", "board meeting on Thursday" — and it names the skill.
Never miss a new skill. New ones drop regularly — subscribe to the newsletter and get a short email with a real example whenever they launch. No spam, unsubscribe anytime. Prefer no email? Follow via RSS or browse the newsletter archive.
🧠 Not just what to do — how to think
Most skills here answer "do this task." A new family answers "think differently about my life."
LLMs have one big weakness: they're too correct. On open-ended questions they give the safe, average, textbook answer — technically right and completely forgettable. Two new bundles fight that head-on (inspired by parallel-divergent-ideation research):
It's not just a folder of files — the whole library is explorable, runnable, and a little bit magic. All of this runs in your browser, free, nothing to install:
claude mcp add pm-skills -- npx -y pm-claude-skills-mcp
No npm install needed — npx pm-claude-skills … always runs the latest. npx pm-claude-skills list shows everything in your terminal. Full per-tool instructions: docs/installation.md.
📚 The skills
Every skill follows the same discipline: what it produces, the inputs it needs, a real framework (severity scales, decision rules — not vibes), a concrete output template, quality checks, and anti-patterns. All 1153 pass the SkillSpec L3 gate and a security audit in CI.
…plus HR, sales, operations, research, healthcare, educators, writers, social media, and more — the full profession index, or by bundle in plugins/ (130 bundles). Install any bundle: /plugin install pm-decoders@pm-skills.
Meta
Before installing anyone's skills (including these): skill-vetting — a security read for SKILL.md files. The library's own standard lives in SKILLSPEC.md; every skill's level is enforced in CI.
🔍 What does a skill look like?
A skill is a single markdown file with a name, a description that tells the assistant when to activate it, and a body containing the working framework: required inputs, decision rules or severity scales, a concrete output template, quality checks, and anti-patterns. The assistant reads it and gains the judgment; humans can read, audit, and edit the same file. No runtime, no lock-in.
markdown
---
name: lease-decoder
description: "Decode a residential lease into plain English and rank the
clauses that can hurt you. Use when someone asks 'what am I signing'…"
---## Framework: Severity Scale- 🔴 Can cost you real money — auto-renewal into a full new term, break
penalties beyond re-rental costs, deposit conditions written to fail…
That's the whole trick: it's markdown. Your agent reads it and gains the judgment; you can read it too, audit it, edit it, or write your own. No lock-in, no runtime, no telemetry.
💸 What it costs you, and how to prove it
Cut your token bill
The pm-tokens bundle optimizes every stage of your agent's token journey — no API keys, stdlib Python, nothing leaves your machine. Five habits, typically 30–60% off a session's token flow:
bash
# 1. Map the repo instead of reading it (~3% of the cost of reading everything)
python3 skills/repo-map/scripts/repo_map.py .
# 2. Crush bulk before it enters context (98% smaller on uniform JSON; errors always survive)
python3 skills/context-crusher/scripts/context_crush.py --mode json --file response.json
# 3. Measure what anything costs — at YOUR prices, times YOUR call volume
python3 skills/token-cost/scripts/token_cost.py --file CLAUDE.md --price-in 3 --calls 200
Plus the judgment skills: token-diet (output costs 3–5× input — diet it where safe), context-budget (cache-aware layout: stable first, volatile last), and session-handoff (resume at ~5% of transcript size). See your own breakdown in the 🪙 Token Dashboard — paste what rides in your context, get computed per-piece savings, all in-browser. The full how-to: docs/SAVE-TOKENS.md.
🤝 Make the most of the cowork skills
The pm-cowork bundle is 100 skills for the office work an AI coworker actually does. Install it (/plugin install pm-cowork@pm-skills), then — the whole trick — describe your mess, don't name the skill: say "my inbox is 4,000 deep", "nobody reads my status updates", "this spreadsheet came from someone who left" — the right skill activates on the ask.
Start where it hurts:
Your pain
Say this
The skill that answers
Drowning in email
"triage my inbox and cut the volume at the source"
Three habits that compound: (1) The weekly review is the keystone — it feeds task-triage-matrix, deep-work-blocking, and personal-wip-limits automatically. (2) The skills chain on purpose — email-to-tasks feeds the task triage; the meeting audit feeds async-instead; delegation-brief hands off what the triage says to shed — follow the links inside each skill. (3) Teams adopt one norm at a time — start with agenda-or-cancel or working-agreements, let it stick, then add the next; the ten-norms-on-Monday rollout is how none of them survive.
Prove a skill works, and stop paying MCP rent
Two CLI tools for the trust-and-cost problems the ecosystem keeps hand-waving — both keyless-to-inspect, both one command:
bash
# Does your skill actually work? Prove it. Paired A/B — skill on vs off, same tasks,# REAL token counts from the API's usage fields, optional blind judge, sha-pinned receipt.
npx pm-claude-skills prove --skill ./my-skill --tasks tasks.txt --runs 2 --judge
npx pm-claude-skills prove --skill ./my-skill --tasks tasks.txt --dry-run # plan + call count, spends nothing# Your MCP servers are charging you rent. Measure it: per-server token cost,# unused-in-N-days flags, "disconnect these three, save X tokens per message".
npx pm-claude-skills mcp-audit --connect
prove exists because the ecosystem is full of "65% better!" claims and almost none are measured — it's the honest-broker harness (the JetBrains "advertised 65%, measured 8.5%" story is exactly why). mcp-audit reads your Claude configs, speaks real MCP to each server to count its schema tokens, and scans your session logs for what you actually use. See also the 📊 AI Spend page — every agent's cost (Claude Code, Codex, Copilot) in one meter, all in-browser.
Agent safety: the pm-seatbelt bundle is the pre-flight checklist before an agent touches email, the browser, or files — least-privilege reviews, prompt-injection spotting, and the blast-radius drill for going autonomous. And RFC 0002 — HANDOFF.md is a dead-simple session-handoff convention (your agent, but it remembers Monday) — a file, not a server, with reference hooks.
Quality, not just quantity
Every skill passes the SkillSpec L3 gate — structure, framework, quality checks, anti-patterns — enforced in CI on every commit
Security-audited — a dedicated CI workflow sweeps every skill and script; calculators are stdlib-only and deterministic with byte-exact output tests
Honest by design — decoders end with a not-legal-advice line, calculators name what they don't model, simulators debrief out of character, and skills that shouldn't ghostwrite (student statements) coach instead
🎁 Beyond the skills (the bonus material)
The library grew an ecosystem — all optional, all linked from the full showcase:
The validator that keeps these 1,099 honest, as a GitHub Action:
yaml
-uses:mohitagw15856/pm-claude-skills@v76with:path:.claude/skills# optional — it finds them otherwise
It checks frontmatter, the Use when … trigger clause a model actually matches
on, leftover template text, and structure — and annotates each finding inline
on the pull request diff, because a finding on the line beats a finding in a
log nobody opens. Also available as npx pm-claude-skills skillcheck.
Zero dependencies, no Docker image, no model call.
Companion tools — for the bits a skill shouldn't guess
A skill can tell a model to check the contrast. Only arithmetic can actually
check it. Where a question has a right answer rather than a good one, the
skill calls out to a tool instead of estimating — both are MIT, zero-dependency,
and neither makes a model call, so they cost nothing to run and return the
same answer every time.
notugly — design systems that are
provably not ugly. #777777 on white is 4.478 and fails AA; #767676 is 4.542
and passes, and no amount of looking at a screenshot separates those.
accessibility-audit, design-system-audit, design-handoff-brief,
brand-guidelines and the Figma reviews now fill their contrast rows from
npx notugly; the MCP server exposes check_contrast directly; and
design-system-generate wraps it for the case
where there is no design system and something ships on Thursday.
rulebook — 37 games, 203
rulings, and how commonly each house rule is actually played.
board-game-night-planner uses it for teach
times and for settling the argument, because a rules disagreement is usually two
groups who learned it differently and are both partly right.
🆕 Latest
v76.2.1 — SkillCheck as a GitHub Action, and the design skills now compute their contrast numbers instead of estimating them.
Everything else is in the changelog and the
releases — a
README should say what this is, not what it was.
❓ First-timer questions, straight answers
Is it actually free?
Yes — MIT, all 1153 skills, forever. The skills are markdown; there is nothing to gate. Sponsors fund the playground's free model runs, not access.
Do I need an API key?
Not to browse, read, install, or use skills inside a tool you already have (Claude Code, ChatGPT, Cursor…). The playground even serves a few sponsor-funded free runs a day. A key only enters the picture for optional extras like running skills from CI.
I'm not a product manager. Is this for me?
PM stands for Professional here. Most of the library is decoders for leases and medical bills, salary-negotiation practice, career-moment kits, life admin, and 35 professions from teaching to veterinary. The product-management corner is just where it started.
Will this mess with my existing setup?
No. Skills are inert text files in a folder; your assistant reads them when relevant. Remove the folder and it's like they were never there. The CLI never touches anything outside the skills directory it tells you about.
How do I know these are any good?
Every skill passes a structural gate (SkillSpec L3) and a security scan in CI; 208 outputs are eval-scored in the open (avg 4.8/5), and the benchmark report publishes the negative findings too. When something's machine-translated or unscored, it's labelled.
MIT — use them, fork them, ship them at work. Skills are judgment, and judgment wants to be free.
Built by Mohit with Claude. 1153 skills · 130 bundles · 35 professions · every commit gated. The long version of this README — every feature, wave, and frontier bet — lives in the Showcase.
