HelloPM · AI Builder Series Explore builds →
Quest 7 · Your own PM AI Agent

Your own PM AI Agent

AI agentsGitHubPM toolingGitHub repo

The assignment

Fork Hermes or OpenClaw on GitHub and ship your own Product Manager AI Agent.

Screenshot of Shiv Kumar's Your own PM AI Agent submission
Your own PM AI Agent — Quest 7, built by Shiv Kumar

What Shiv shipped

Atlas is a Product Manager AI copilot persona defined in a SOUL.md configuration file, built using RICE and JTBD frameworks with an evidence-honesty rule set. It includes a skills folder, an install script, and a sample PRD (streak-freeze-prd.md) demonstrating its output.

Open the GitHub repo ↗ Shiv's full portfolio

Our read on this build

Shiv shipped a public GitHub repo. This read comes from the artifact and Shiv's own account of it — no scores, just what the build does and does not show.

What it demonstrates

  • It stands on a real codebase — agent config files and a public repository — which is why it exists after a week instead of a quarter.
  • It is bent to specific PM work: PRD generation, a defined persona and custom skills — a workflow, not a rename.
  • And you can watch it work: a demo.

Take this awayLook at how it is decomposed — named skills for named jobs. That structure is why you can trust an agent with one workflow instead of hoping across ten.


Why this quest matters

Forking beats a blank page. Standing on Hermes or OpenClaw and bending it to a real PM workflow is exactly how modern builders ship a working agent in days instead of months.

What separated the strongest builds

Across the 24 builds we could open for this quest, the same three or four things kept deciding whether one landed. Useful whether you are judging this one or attempting your own.

It stands on a real codebase

A linkable fork of Hermes or OpenClaw. 'Inspired by' is how people describe a blank page they started anyway.

Bent to one PM job, properly

A PRD writer, a backlog triager, a teardown engine — a workflow the builder actually has. A renamed system prompt on a template is not customisation.

Seen doing the work

Running, answering, producing an artifact you can look at. A repo that claims to work and is never shown working asks for a lot of trust.

No numeric scores are published here. These pages are written reads of what each builder shipped — the build is the point, not a mark out of a hundred.

More from Shiv

Quest 1Quest 2Quest 3Quest 4Quest 5Quest 6Quest 7
← Back to Shiv's portfolio