bpm
Behavioral Package Manager
Install conventions, extensions, and templates for your AI agent
The package manager for the behavioral layer that AGENTS.md was built to carry
A shared layer for defining AI agent behavior
bpm registry creates a place where behavioral packages for AI agents can be discovered, installed, and improved in the open. It turns hard-to-transfer workflow habits into reusable conventions, templates, extensions, and measurements.
The problem
Without behavioral packages, agent behavior is unpredictable across runs and collaborators. Quality and consistency drift when assumptions live in ad-hoc prompts instead of versioned, shared configuration.
What bpm changes
Behavioral packages make agent behavior reproducible. Anyone can define how an agent should work - for their team, their field, their workflow - and share it as a versioned, installable package.
Why contributions matter now
Agent behavior improves when more perspectives shape it. Every published package - from any discipline - expands what's possible for everyone building with agents.
Ecosystem flywheel
Anyone publishes conventions for their field → builders install them → agents become more predictable → more perspectives join → the ecosystem gets smarter.
Why use bpm?
Mirror
Conventions shape behavior. Install shared patterns that make your agent work the way your team works.
Compose
Packages work together. Layer conventions, extensions, and templates into a coherent agent configuration.
Measure
Track what's working. Measurement packages give you visibility into agent performance and cost.
Quick Start
Install the bpm CLI globally via npm to get started.
Requires Node.js 18+. After install, run bpm --version to verify.
Quick Start - Choose a Flow
Flow A - Contributing and Publishing
# Stage 1A - existing package repository
$ cd /path/to/package-repo
$ bpm init
# Stage 1B - new package repository
$ cd ~/your-workspace
$ bpm init my-bpm-package --from-scratch
$ cd my-bpm-package
# Stage 2 - publish
$ bpm token set <token>
$ bpm publish --auth-source=stored
Flow B - Install into current agent workspace
# existing project or global agent workspace with AGENTS.md
$ cd /path/to/workspace
$ bpm init
$ bpm install learning-mode
$ bpm list
# AGENTS.md managed section is updated
Featured Packages
Featuredgoose-banner
Customizable session startup banner - drop scripts in banner.d/ to surface context at session open
learning-mode
Shifts agent posture from delegation to scaffolding for skills at DOK 1-2
open-agent-harness
Multi-agent testing harness - evaluate agent behavior across scenarios and configurations