PM Skills — Professional Agent Skill Library
1078 professionally-crafted, plain-markdown skills that teach any AI assistant to perform senior-level tasks — from PRDs to lease decoding.
Evidence shows: skill files are plain text, no runtime, no telemetry, no accounts; installation copies text files, uninstall deletes them; security policy explicitly states no content collection, no key transmission; CI includes security audit and SkillCheck; dependencies minimal (only devDependency qrcode), no third-party runtime dependencies; external effects limited (e.g., MCP servers, Telegram bot, all open-source and user-configured); rollback supported via lockfile and verification; source attribution clear (MIT license, author info). Deductions: no actual runtime verification, and some external integrations (e.g., hosted MCP) trust chain relies on claims not verified in static review.
Evidence shows: README and SECURITY.md consistent, skill structure uniform (SKILL.md format), CI checks generated artifacts consistency; dependency availability high (pure stdlib scripts, no third-party deps); but failure messages only seen in CI error output, no user-facing error handling docs. Deductions: no actual runtime verification, failure message evidence thin.
Evidence shows: broad audience (individuals, professionals, students), wide scenario coverage (from PRDs to life admin); capability boundaries clearly stated in SECURITY.md (no code execution, no filesystem access, etc.); trigger methods diverse (natural language, commands, plugins); environment fit broad (Claude Code, ChatGPT, Gemini, Cursor, Codex). Deductions: some capability boundaries (e.g., MCP servers) rely on external services not verified in static review.
Evidence shows: clear information architecture (directories, categories, SKILLS.md catalog); detailed install notes (multiple tools); naming stable (consistent skill names); examples and FAQ abundant (many examples in README); known limitations not explicitly listed (only partially in security policy); MIT license clear; version number exists (package.json 75.1.0), but no CHANGELOG; maintenance responsibility clear (author, contributing guide, security reporting channel). Deductions: known limitations evidence thin, versioning changelog missing.
Evidence shows: high output usability (templates, quality checks, anti-patterns); high marginal value (1078 skills, broad coverage); good cost-benefit (free, open-source, no runtime). Deductions: no actual runtime verification of output quality, but static evidence strong.
Evidence shows: README claims (e.g., skill count, SkillSpec pass) supported by corresponding files (SKILLS.md, conformance/REGISTRY.md); cross-source corroboration (Anthropic plugin directory, npm, PyPI); fact/inference separation (README distinguishes descriptions from claims). Deductions: some claims (e.g., security audit) rely on CI config not actually run.
- Static review cannot verify actual runtime behavior; all security claims are based on documentation and CI config.
- Trust chain for external integrations (e.g., hosted MCP, Telegram bot) relies on claims not verified in static review.
- No explicit known limitations document found; users should assess skill applicability themselves.
What does this agent do, and when should you use it?
PM Skills is an open-source library of 1078 Agent Skills — each is a SKILL.md file that teaches an AI assistant to perform one professional task to a senior professional's standard, from writing a PRD to decoding a lease or running a blameless postmortem. Every skill bundles a framework, output template, quality checks, and anti-patterns. It's MIT-licensed and works with Claude, ChatGPT, Gemini, Cursor, and Codex. Installation involves copying text files; there's no runtime, telemetry, or lock-in.
The library provides 1078 skill files, each a plain-markdown file with a name, a description for when to activate, and a body containing required inputs, decision rules/severity scales, an output template, quality checks, and anti-patterns. Users can install via npx pm-claude-skills add or /plugin in Claude Code. Skills can be run in the browser-based Playground (with limited free runs), and are also accessible via an MCP server and a REST API. Some skills include Python scripts (e.g., for token-cost calculation). All skills pass SkillSpec and a security audit in CI.
- A renter decodes a lease before signing, using the lease-decoder skill to surface risky clauses.
- A product manager drafts a PRD with prd-template and prioritizes backlog with rice-prioritisation.
- A job seeker rehearses salary negotiation with salary-negotiation or the promotion committee simulator.
- A new parent organizes logistics with new-baby-logistics.
- Someone feeling overwhelmed uses where-do-i-start to identify their next concrete action.
- A developer cuts token usage with token-diet and context-crusher to reduce context overhead.
What are this agent's strengths and limitations?
- 1078 skills covering 35 professions and life situations, regularly updated
- All skills are plain markdown — easy to audit, edit, and review, with no vendor lock-in
- Skills are gated by a structural spec and security audit, with public eval scores (avg 4.8/5)
- Works with multiple AI tools: Claude Code, ChatGPT, Gemini, Cursor, and Codex
- Large catalog may require time to find the right skill, though searchable catalog and find page help
- Some skills depend on Python scripts, requiring a Python 3 environment
- Playground free runs are limited; heavy use may require your own API key or sponsor support
- Quality varies by contribution; not all edge cases may be covered — users should exercise judgment
How do you install or deploy this agent?
For Claude Code: type /plugin and search 'pm-skills' (published in the official Anthropic directory), or run npx pm-claude-skills add --agent claude. For other tools (Cursor, Codex, Windsurf, Cline...): run npx pm-claude-skills add --agent cursor (or the relevant tool name). No npm install needed; npx always runs the latest.
How do you use this agent?
After installation, simply ask in plain language, e.g., 'decode this lease' or 'write a PRD for our referral feature.' The AI assistant reads the relevant SKILL.md and follows its framework. You can also browse the full catalog in SKILLS.md or run skills in the browser Playground without installing anything.