PM Skills: 1,153 Markdown Workflows That Turn an LLM Into a Senior Professional

What PM Skills Actually Solves

Generic AI assistants give you a plausible answer for almost anything, but the answer is often shallow. PM Skills addresses that by providing 1,153 agent skills, each one a plain-markdown SKILL.md file that teaches an AI to perform one specific task to a senior professional's standard. The tasks range from writing a PRD to decoding a lease, running a blameless postmortem, or preparing a salary negotiation. The target user is someone who wants the AI to follow a structured framework, not just free-associate. The README explicitly positions this as a tool for moments where filler is not acceptable: a landlord keeping a deposit, a confusing medical bill, a layoff, or a death in the family. It is for professionals who need a checklist and a template, not for tinkerers who want to build their own prompt chains from scratch.

The Mechanism: A Skill Is a Markdown File, Nothing More

The core design is simple: each skill is a SKILL.md file. The README states that a skill is a markdown file your AI reads, with no runtime, no telemetry, no accounts. Each skill bundles a framework, an output template, quality checks, and anti-patterns. This means the AI reads the file, follows the instructions, and produces an artifact. The installation process is just copying text files; uninstalling is deleting them. This is a deliberate choice to keep the barrier low. There is no plugin runtime, no API, no background process. The AI's native ability to follow instructions does the work. The repository includes a skills directory, and the README points to a specific example, skills/lease-decoder/SKILL.md, which it says is designed to be read by humans too. That transparency is a strength: you can read the exact instructions your AI will follow before you trust it with a sensitive task.

Getting It Running: Three Paths, All Low Friction

The README offers three installation routes. First, for Claude Code users, you type /plugin, search for pm-skills, and install. No command line needed. Second, for everyone else, you run npx pm-claude-skills add and pick your tool from a menu that includes Cursor, Codex, Windsurf, ChatGPT, and Gemini. Third, there is a browser playground at the project's GitHub Pages site where you can run a skill without installing anything. The playground lets you pick a skill, fill a short form, and stream out an artifact. There is also a find.html page where you describe your task in plain words, like "my landlord kept my deposit," and it names the skill. The README emphasizes that each door takes about 30 seconds. For those who want to browse, there is a separate catalogue site at site-jet-seven-34.vercel.app with a newsletter and RSS feed for new skill announcements.

Beyond Task Skills: The Thinking and Focus Bundles

Most skills answer the question "do this task." The README introduces two additional families that answer "think differently about my life." The pm-thinking plugin includes skills like the-third-answer, five-minds, decision-panel, and red-team-my-plan. These are designed to counter the tendency of LLMs to give the safe, average, textbook answer. For example, devils-advocate-on-demand fights the "that's great!" bias, and poke-holes-in-this forces the AI to list only weaknesses. The pm-focus plugin is aimed at executive function, with skills like where-do-i-start, task-to-first-step, and overwhelm-triage. These are described as ADHD-friendly but useful for everyone. This is a meaningful expansion of the library's scope: it is not just about producing documents, but about changing how the AI approaches open-ended questions. The README cites parallel-divergent-ideation research as inspiration, though it does not link to a specific study.

A Real Limitation: Volume Over Curation

With 1,153 skills, the library's biggest risk is variable quality. The README does not describe a formal review process for each skill. It mentions a conformance/REGISTRY.md and GitHub Actions workflows named skillcheck.yml and skill-audit.yml, which suggest some automated checks exist, but there is no evidence of human expert review for every skill. For high-stakes tasks like decoding a lease or handling a medical bill, an incorrect or outdated framework could cause real harm. The README itself acknowledges skepticism by telling you to read a skill first, which is honest but also a sign that the maintainers expect you to do your own due diligence. Another limitation is that the skills are only as good as the underlying model's instruction-following ability. If the AI misreads the markdown or skips a quality check, the output could be incomplete. The library does not guarantee a specific output format beyond what the skill file describes.

Alternatives: Comparing to Built-In Prompts and Other Skill Libraries

The most direct alternative is to write your own prompts or use the built-in system prompts of tools like ChatGPT or Claude. That approach gives you full control but requires time and expertise to encode a senior professional's framework. PM Skills offers a shortcut: pre-written, structured instructions. Another alternative is the broader ecosystem of agent skill libraries, such as the awesome-claude-skills list that the README links to. Those libraries aggregate skills from multiple authors, but they are not a single curated collection with a consistent format. PM Skills differentiates itself by having a unified structure (SKILL.md with framework, template, quality checks, anti-patterns) and a single installation command. The trade-off is that you are tied to this project's conventions and its maintainers' judgment on what constitutes a senior standard. There is also a Python package on PyPI (pm-skills) and a remote MCP server (mcp-remote/), which suggests the skills can be exposed via the Model Context Protocol, but the README does not explain how to use those.

Maintenance and Licensing: Active but Unclear Governance

The repository is MIT-licensed, which means you can use, modify, and redistribute the skills freely, even commercially, as long as you preserve the copyright notice. The README does not mention any contributor guidelines or a code of conduct. The release history shows frequent updates: v77.0.0 on 2026-08-18, v76.2.1 on 2026-08-16, and v76.2.0 on the same day. This suggests an active maintenance cadence, but the version numbers are high (77.0.0), which could indicate many incremental changes or a versioning scheme that is not semantic. The last push date is 2026-08-18, so the project is current as of this writing. There is a GitHub Actions workflow named skill-audit.yml, which likely runs checks on the skills, but the README does not describe what the audit covers. For upgrade cost, the npx command likely fetches the latest version, and since skills are just files, updating is a matter of re-running the install command. There is no mention of a migration path for custom modifications, so if you edit a skill file locally, an update might overwrite your changes.

Editorial conclusion

Adopt PM Skills if you want a large, structured library of professional frameworks without building your own prompts from scratch, and if you are comfortable with plain markdown files as the delivery mechanism. Do not adopt it if you need deep integration with a specific vendor's proprietary skill system, or if you expect the library to be a polished product rather than a community-driven collection. Before relying on it for a high-stakes task like a lease or a medical bill, read the actual SKILL.md file for that skill to verify the quality and accuracy of the framework it contains. The repository is actively maintained with frequent releases, but the sheer volume of skills means individual quality will vary, so spot-check the ones you plan to use.

Official sources

1. Official documentation
2. Official README
3. Project repository
4. Release notes