Bear
AI-driven package manager for Skills, Rules & Memory
Start new thread

Bear - AI-driven package manager for Skills, Rules & Memory

by•
CTXPM (Context Package Manager) is an open-source, AI-first protocol that organizes Skills, Rules, Prompts, Memory, and MCP into two clear ownership types - dependency and package - so your AI coding agent can check, update, and validate them on its own. * dependency: external AI resources (Git, OCI, file path) you don't own. Tracked by source/version, not committed to your repo. * package: resources your project owns. Reviewed and versioned right alongside your source code.

Add a comment

Replies

Best
Maker
📌

👋 Hey Product Hunt!

I built Bear. CTXPM because every AI coding project I worked on eventually became a junk drawer: skills, rules, prompts, MCP configs, and memories piling up with no clear line between "stuff I imported" and "stuff my project actually owns." Months later nobody could tell which files belonged in Git, whether an external skill was outdated, or whether a rule applied to the whole team or just one repo.

CTXPM (Context Package Manager) is an open-source protocol and CLI that treats AI resources like code dependencies, with two semantics:

  • dependency: external Skills/Rules/Prompts/Memory/MCP you don't own, tracked by source and version, not committed

  • package: resources your project owns, reviewed and versioned with your source code

The protocol is AI-first. The rules live inside your agent's entry doc ( / / etc.), so your coding agent reads them itself and autonomously checks for dependency updates, fixes broken entry symlinks, and flags unmanaged resources during normal coding and review. It only interrupts you for real decisions: migrate this? upgrade that? which rule wins a conflict?

Getting started is one line you hand to your agent:

Please detect, install, initialize, and transform this project according to the instructions in  to make it a CTXPM protocol structure for managing AI dependencies and AI resources.

100% open source:

Would love feedback, especially from anyone juggling multiple agents (Claude Code, Cursor, etc.) across the same repo, or maintaining shared skills/rules across many projects. AMA! 🐻‍❄️