Angry Bear
Angry Bear is the enforcement layer we built so our coding agents load the right context before they touch a file. It is a pre-tool-use hook that requires the right skill to be loaded before a given tool can edit a given path - turning 'please follow the conventions' into a precondition. Works with Claude Code and Cursor.
Angry Bear is the enforcement layer we built so our coding agents load the right context before they touch a file. It is a pre-tool-use hook that requires the right skill to be loaded before a given tool can edit a given path - turning "please follow the conventions" into a precondition. Works with Claude Code and Cursor.
Skills give a coding agent the conventions and context a task needs; the trouble is getting the agent to load the right one at the right moment, every time, even after the conversation compacts. Angry Bear closes that gap. It sits in front of the agent’s tool calls and holds an edit until the skill that governs that file has been loaded in the session - so the guidance is present at the moment of editing, not 40k tokens ago.
How It Works
Angry Bear runs as a pre-tool-use hook. Rules are declarative and keyed on (tool, path, agent) → required skill: edit a Go file, and go-coding-standards must be loaded first; touch a test, and testing-architect must be loaded. When a required skill is missing, the hook returns a short message telling the agent which skill to load, the agent loads it, and the retry proceeds - now with the right conventions in context.
Declarative Rules Engine
A small, code-free JSON policy keyed on tool, path glob, and agent. Add a row to require a skill for a file pattern; there is nothing to compile.
Agent Adapters
Works with Claude Code and Cursor out of the box. Anything else integrates by implementing a single adapter interface - no lock-in to a specific agent.
Per-Machine or Per-Repo
Keep rules local to your machine, or commit them to the repository so the whole team inherits the same gate on the same branch.
TUI Dashboard
A terminal dashboard shows your skills, rules, and a live event log of every load and every gated tool call, so the policy is observable rather than opaque.
Why It Is Open Source
Enforcement only helps if you control it. Angry Bear is a local hook with no service to call and no vendor lock-in: you bring your own skills, your own rules, and your own agent. It is open source under the MIT license so any team can drop it in front of their coding agents and get the same discipline we rely on internally.
Read the launch post for the story behind it, or go straight to GitHub for install and setup.