Key Takeaways

  • Ezra Mechaber, Jinan, and Lawrence created PM OS as a shared Git repository containing 31 skills for Figma's 60 to 70 product managers.
  • The architecture uses two Git repositories: a shared parent repo for global agent instructions and a private sub-repo where PMs save notes securely.
  • Figma PMs execute skills inside Claude Code and terminal environments instead of relying on a standalone SaaS product.
  • Mechaber runs a custom skill called "what's up" six times a day to aggregate Slack noise, track active deliverables, and flag unresolved decisions.
  • Specialized internal modules automate downstream technical tasks, including AutoXP for experimentation and Codex for QA bug bashes.

The Method

Mechaber started what became PM OS alone at his desk. “I started thinking about like what would it look like to turn this thing that I was doing just like hunched over my laptop like a little creep into this kind of reusable thing that lots of other PMs could build on top of,” (play from 4:11) Mechaber said. Collaborating with fellow PMs Jinan and Lawrence, he turned personal command line scripts into a shared operational system for 60 to 70 product managers.

The system relies on two technical choices:

First, the dual-repo pattern solves data privacy without custom server infrastructure. “It's technically two get repos. It's the parent repo that everyone can download and then their workspace is a sub repo so that if you want to write stuff back, you can write it back privately,” (play from 11:39) Mechaber explained. The parent repo includes an agent.md file that sets behavioral rules, alongside 31 individual skill files that define trigger events and tasks. The child workspace repo lives on each PM's machine.

Second, the system lives directly in the terminal via Claude Code. Instead of opening third-party tools, PMs execute workflows in their shell. For administrative context, skills crawl internal codebases to locate code owners and populate reviewer lists on product documents. Claire Vo noted the value here: “In large organization, it is actually really hard to know like who should have an opinion on this product? Is somebody else working on it? And so, I like that you invested in sort of like the baseline context gathering as skills.” (play from 9:43)

Daily communication triage is similarly automated. Mechaber runs a command called "what's up" up to six times every day: "It pulls all of my relevant slacks and projects together. It pulls all of the things that I like who I owe stuff to, what decisions are outstanding." For technical product operations, Figma built specialized modules: AutoXP handles autonomous multi-agent experimentation pipelines, Jev checks product writing, and Codex runs automated QA bug bashes.

Where This Breaks Down

Terminal workflows work only when product managers are comfortable inside Git repositories and command lines. If an organization hires PMs without basic technical fluency, adoption stops immediately.

Dual-repository setups also suffer from drift. When core maintainers update skill files in the parent repo, local clones can fall out of sync or miss critical agent instructions unless team members actively pull upstream changes. Finally, context gathering across Slack and code repos burns prompt tokens fast. Without strict boundaries on message history and channel scope, automated queries will overload context windows with irrelevant chatter and generate high API bills.

What to Do With This

Audit your team's weekly schedule and identify the administrative tasks that consume 30 minutes every morning. Create a bare Git repository with an agent.md file defining your team's document templates, reviewer assignments, and project slack channels. Test running single-line status queries in Claude Code with one technical PM before attempting to distribute shared skills to the wider team.