Shared infrastructure for alignment researchers. MATS fellows are the first intended users, but this is built for the broader AI safety community.
Large orgs benefit from shared tooling and accumulated knowledge across their agents. This project aims to bring some of those advantages to independent researchers:
- Plugin marketplace - Curated Claude Code plugins with skills for common research workflows
- hive-mind - A system for sharing session learnings across the community (in development)
Install Claude Code if you haven't already:
curl -fsSL https://claude.ai/install.sh | bashRun the following commands from within Claude Code (the / prefix indicates a Claude Code command):
/plugin marketplace add Crazytieguy/alignment-hive
Enable auto-update to get new plugins and updates automatically:
- Run
/plugin - Go to the Marketplaces tab
- Select alignment-hive
- Select Enable auto-update
- Press Esc twice to exit the menu
Install the mats plugin (recommended for MATS fellows):
/plugin install mats@alignment-hive
The mats plugin includes:
- best-practices - Guided setup for documentation, plugins, tooling, and permissions
- fellow-handbook - Quick lookup of MATS policies, compute access, housing, and program logistics
- permissions - Configure Claude Code permissions to reduce permission prompts
- lit-review - Automated literature review with paper search and summarization
Also available:
/plugin install llms-fetch-mcp@alignment-hive
Adds an MCP server for fetching documentation with llms.txt support.
Exit Claude Code and navigate to your project directory:
/exit
cd ~/my-project && claudeWorks for both new and existing projects. Ask Claude to help you set it up!
The plugin-dev plugin auto-installs when you clone this repo, so Claude can help with plugin development.
Feedback and suggestions welcome—open an issue, send a Slack DM, or reach out however works for you. All changes go through PR review.
A web interface for hive-mind is in development at alignment-hive.com.
See docs/roadmap.md for what's planned.