pm-brain  by phuryn

A second brain for product managers, built on local markdown

Created 1 month ago
477 stars

Top 63.3% on SourcePulse

GitHubView on GitHub
Project Summary

PM Brain OS addresses the challenge of scattered product management context by providing a "second brain" solution. It centralizes knowledge, decisions, and hypotheses in plain Markdown files locally, integrated with Claude Code. This aims to improve recall, maintain context, and ensure traceable decisions for PMs overwhelmed by fragmented information.

How It Works

The system uses a core loop: ingest artifacts, synthesize, propagate updates to structured stores (knowledge/, hypotheses/, decisions/, stakeholders/), and tag claims with provenance. It avoids vector databases, cloud storage, and agent memory tricks, relying on grep-able Markdown. A weekly /review sweep flags drifting information and prompts actions, keeping the knowledge base current.

Quick Start & Requirements

  • Primary install / run command:
    • Install skill: curl (macOS/Linux/WSL/Git Bash) or powershell (Windows).
    • Bootstrap brain: cd <folder>; claude then /pm-brain.
  • Prerequisites: Claude Code, Git.
  • Links: docs/walkthrough.md, docs/how-it-works.md.

Highlighted Details

  • Commands: /ingest, /prep, /review, /ideate, /risk, /plan streamline PM workflows.
  • Provenance tagging: Claims are tagged with source (documented, verbal, hunch) for audibility and override.
  • Operates locally with plain Markdown, ensuring data accessibility and auditability without external services.
  • Weekly /review command synthesizes state, flags issues, and drafts action items.

Maintenance & Community

  • Designed and maintained by Paweł Huryn.
  • Described as a "Research preview" with recent organizational installs.
  • Contributions via GitHub issues encouraged, with specific guidance for skill vs. documentation changes.

Licensing & Compatibility

  • License: MIT.
  • Compatibility: Permissive for commercial use and integration.

Limitations & Caveats

A "Research preview" with very recent real-world deployment. Not for autonomous PM; user judgment is key. Evaluation suite reports 404/406 checks passing (≈99.5%), with two known residual LLM-judge failures on long scenarios.

Health Check
Last Commit

1 month ago

Responsiveness

Inactive

Pull Requests (30d)
1
Issues (30d)
1
Star History
153 stars in the last 30 days

Explore Similar Projects

Feedback? Help us improve.