What is an AI PM operating system? The options compared
An AI PM operating system is a set of files your AI agent reads before it answers: context about your company and product, reusable PM skills, chained workflows, and review agents that critique the output. It runs inside an agent like Claude Code, Codex, Claude Cowork or Cursor, so every draft starts from your situation instead of a blank prompt. You can assemble one from free GitHub skill packs, take a course and build your own, or buy a ready-made one.
Disclosure: I make AI PM OS, one of the options below. I've described the free options using their makers' own published facts, checked on 24 October 2026.
What an AI PM operating system is (and isn't)
A PM OS is a folder. Your agent reads it, and it changes what the agent knows and how it works before you type anything.
It is:
- Context files: who your company sells to, what the product does, your team, this quarter's goals, the constraints you can't break.
- Skills: single PM tasks the agent can run, such as a JTBD interview guide or a RICE scoring pass.
- Workflows: several skills chained in order, such as draft a PRD, review it, revise it.
- Review agents: personas that attack a draft before your stakeholders do.
- Memory: decisions and risks captured as you work, so next week's session knows what happened this week.
It isn't:
- A SaaS dashboard. Everything is Markdown on your machine.
- A prompt library. Prompts are copied in by hand; skills are picked up by the agent when they fit the task.
- A work tracker. Linear or Jira hold the tickets. A PM OS encodes how the thinking runs. I wrote up that split in AI PM OS vs Linear MCP.
The five layers of a PM OS
Every serious setup I've seen, paid or homemade, ends up with the same five layers. The names differ; the jobs don't.
1.Context
Five short files cover most of it: company, product, team, goals, constraints. This layer is what stops the "plausible but generic" PRD. A model that doesn't know you're B2B, that mobile was cut last month, or that legal blocks storing EU data will guess, and its guess is the average of the internet. The AGENTS.md guide shows how to write the file your agent loads first.
2.Skills
A skill is one PM task written down well enough that an agent can run it: the steps, the questions to ask, the output format, the mistakes to avoid. In Claude Code a skill is a folder with a
SKILL.mdfile, and the agent loads it when the task matches (Anthropic's skills docs). How to organise Claude Code skills covers structure.3.Workflows
Real PM work is rarely one step. A workflow chains skills so the output of one becomes the input of the next. Example: research transcripts → cleaned notes → JTBD extraction → insight clusters → a hypothesis → an experiment design.
4.Review agents
A review agent is a persona with one job: find what's wrong. Typical ones are engineering (feasibility, effort), executive (so what?), devil's advocate (the weakest assumption) and customer voice (would a user care?). Without this layer, the agent agrees with you.
5.Memory
When you say "leadership killed mobile and moved us to retention", the system should record a decision, a stakeholder and a risk, and load them next time. Good memory asks before it writes to your files.
Skill, framework, workflow: the short definitions. A skill is a single AI-invocable PM task. A framework is the thinking model a skill applies, such as RICE or Kano. A workflow is a chained multi-step process that runs several skills in order.
Do you need one?
Three signs you do:
- You paste the same company background into every new chat.
- Your PRDs and strategy memos come back sounding like anyone's.
- Nobody reviews your work before your eng lead or VP does, and they find the gaps.
Two signs you don't:
- You use AI for quick rewrites and summaries. A chat app is faster for that.
- You won't open Claude Code, Codex, Cowork or Cursor. A PM OS lives in those tools; without them it's a folder of text files.
Build, assemble or buy
Build your own
You write the context files, skills and workflows yourself. Free guides exist: ccforpms.com is a free course taught inside Claude Code, and vishalmdi/ai-native-pm-os is a free 11-module guide on GitHub. Paid cohorts exist too, such as Ron Yang's Build Your PM Operating System on Claude Code on Maven.
You learn the most this way. It costs the most time, and you maintain it.
Assemble from free skill packs
Install one of the free PM packs, then write your own context layer on top. The packs are good. What you still own: picking which skills to keep, writing context files, wiring workflows and connectors, and keeping it current as the tools change.
Buy a packaged OS
Someone else has made the choices and keeps them updated. You get a working setup on day one, you pay, and you live with their opinions about structure.
AI PM operating systems compared (2026)
Counts are what each maker publishes, checked 24 October 2026. Stars change daily.
| Option | Price | License | What ships (maker's numbers) | Context layer | Runs in | Best for |
|---|---|---|---|---|---|---|
| Assemble from free skill packs | ||||||
| Anthropic Product Management plugin | Free | Apache 2.0 (repo) | 7 commands (/write-spec, /roadmap-update, /stakeholder-update, /synthesize-research, /competitive-brief, /metrics-review, /brainstorm), 7 skills, MCP connectors | Pulls from connected tools (Slack, Linear, Notion, Figma and others); otherwise you provide it | Cowork (primary), Claude Code | Teams on Cowork that want official support |
| PM Skills Marketplace (phuryn/pm-skills) | Free | MIT | 69 skills, 42 chained workflows, 9 plugins; ~27.3k stars | You write it | Claude Code, Cowork; skills in Codex, Cursor, Gemini CLI | Breadth and ready slash commands |
| Product-Manager-Skills (Dean Peters) | Free | CC BY-NC-SA 4.0 | 77 skills (v0.84, 10 Aug 2026); ~7.6k stars | You write it | Claude Code, Codex, Cursor, ChatGPT and others | Learning the reasoning behind each framework |
| pm-skills (product-on-purpose) | Free | Apache 2.0 | 68 skills, 10 workflow commands, sub-agents, 200+ sample outputs; ~757 stars | You write it | Claude Code; Cursor, Codex, Copilot via AGENTS.md or skills CLI | Rigor, worked samples |
| Buy a packaged OS | ||||||
| AI PM OS (mine) | $99 one-time | Paid license, 30-day refund | 231 skills, 150+ frameworks, 11 workflows, 12 review agents, memory capture | 5-file context set up with /start | Claude Code, Codex, Cowork, Cursor | Context, workflows and reviewers already wired |
| mySecond | $39 per seat per month, 14-day trial | Subscription | See maker's site | Persistent context (maker's description) | Claude Code | Teams that want a shared, subscribed system |
| Build your own | ||||||
| Build your own (ccforpms, ai-native-pm-os) | Free (your time) | Varies | Whatever you write | You write it | Any agent | Learning by doing |
On licenses. MIT and Apache 2.0 let you use, change and redistribute the files, including commercially. Dean Peters' pack is CC BY-NC-SA 4.0. His README says you can use the skills in your day job and adapt them if you share alike, but you can't repackage them into a paid product. If you plan to build something on top of a pack, read its license first. This is a description, not legal advice.
How to choose
- You want to learn how skills work start with Dean Peters' pack. It's written to teach the why behind each framework.
- Your team is on Cowork and wants the vendor's own plugin Anthropic's Product Management plugin.
- You want the most free commands for PRDs, discovery and strategy phuryn/pm-skills.
- You care about consistent artifacts across a team and like seeing examples product-on-purpose/pm-skills, with 200+ sample outputs.
- You want context, workflows and reviewers working together this week and don't want to maintain them a packaged OS such as AI PM OS.
- You want to build it yourself and understand every file a free course plus your own folder.
Whichever you pick, write your context files. Every option on this list gets better with them and generic without them.
Ready-made AI workflows for PRDs, strategy and research
If what you want is workflows rather than a full system, these exist today:
| Job | Free option | Paid option |
|---|---|---|
| PRD | Anthropic /write-spec; phuryn /write-prd, then /red-team-prd | AI PM OS /workflow:prd: draft → strengths reviewer → flaws reviewer → revision loop |
| Strategy | phuryn /strategy (9-section Product Strategy Canvas) | AI PM OS /workflow:strategy: problem → who it's for → what's unique → competitive position, ending in a strategy kernel |
| Research | Anthropic /synthesize-research; phuryn /discover, /interview | AI PM OS /workflow:research: transcript cleanup → JTBD extraction → insight clustering → hypothesis → experiment design |
For five starter workflows you can run with plain prompts, see AI for product management.
What a PM OS can't do
- It won't make the call. It drafts, structures and challenges. Prioritization, saying no to a VP, and deciding what ships stay with you.
- It inherits your company's AI policy. Context files go to whichever model provider your editor uses, under that provider's terms. If your company restricts AI tools, the OS is restricted too.
- The editor costs extra. Claude Code and Cowork need a paid Claude plan; Codex runs in the ChatGPT desktop app; Cursor has a free tier. See claude.com/pricing.
- Stale context makes it worse. If your goals file still says "grow mobile" after mobile was cut, the drafts will argue for mobile.
Check your setup before you choose
The free AI PM Setup Scorecard grades your current setup across eight dimensions, including context, workflows and quality control, and shows the biggest gaps. Fix those gaps with whichever option above fits.
If you want to see what a packaged skill looks like before spending anything, my free sample is at github.com/gnurio/nurijanian-skills (MIT; start with make-requirements-great). The full system is AI PM OS, $99 one-time with a 30-day refund.
FAQ
What is a PM operating system?
A set of files an AI agent reads before it works: context about your company and product, PM skills, chained workflows, review agents and memory. It runs inside Claude Code, Codex, Cowork or Cursor so drafts start from your situation, not a blank prompt.
Is a PM OS the same as Claude Code?
No. Claude Code is the agent. A PM OS is the folder of context, skills and workflows the agent reads. The same PM OS can run in Claude Code, Codex, Cowork or Cursor, depending on how it's packaged.
Can I build my own PM OS in Claude Code for free?
Yes. Write five context files, add a free skill pack such as phuryn/pm-skills or Dean Peters' Product-Manager-Skills, and save your repeat tasks as skills. Free courses such as ccforpms.com walk you through it. You pay in time and upkeep.
What's the difference between a skill, a framework and a workflow?
A skill is a single AI-invocable PM task. A framework is the thinking model a skill applies, such as RICE. A workflow chains several skills into one multi-step process, such as draft, review and revise a PRD.
Is Linear MCP an AI PM OS?
No. Linear MCP connects your agent to Linear so it can read and update issues. That is where work lives. An AI PM OS encodes how PM thinking runs. They work well together.
Which PM OS works in Cursor or Codex, not just Claude Code?
AI PM OS runs in Claude Code, Codex, Cowork and Cursor. phuryn's skills work in Codex and Cursor, though its slash commands are Claude-specific. Dean Peters' and product-on-purpose's packs both document Codex and Cursor installs.
Do I need to know how to code to use a PM OS?
No. You type plain English in the agent. Installing a pack means pasting one or two commands, and some packs install from Cowork's plugin screen.
Where does my data go?
Your files stay on your machine. When you ask the agent something, the relevant context goes to your editor's model provider under that provider's terms. For AI PM OS specifically, nothing routes through prodmgmt.world.
Free GitHub skill packs vs a paid PM OS: what's the real difference?
The free packs give you well-made skill files. You choose what to keep, write the context files, wire workflows and keep it current. A paid OS ships those parts already assembled and maintains them.