team.management

The model getting more capable does not make it more trustworthy. It makes it a more convincing liar. team.management is an open-source protocol engine for Claude Code that puts every process on rails — enforced as it happens, so your AI can't fake the work.

It installs as a Claude Code plugin. You define a protocol — a JSON config describing a multi-step lifecycle such as investigation, implementation, code review, documentation and completion. Each step declares which tools the agent is allowed to use, which checks must pass, and what it has to produce before it can advance.

Enforcement is mechanical rather than advisory. The plugin registers hooks that run before a tool call is executed, so an agent in a read-only step is blocked from editing files rather than merely asked not to. Steps, task state and git branches are managed by the engine, and every run leaves a work log you can read afterwards — instead of a summary the model wrote about itself.

Protocols included: task, brainstorm, research, refactoring, and two metric-driven optimization loops. You can write your own. Reviews can run across several models — Claude, plus OpenAI Codex and Google Antigravity when you enable them — so a second, independently trained model checks the first.

What it does not do: it will not make an AI write better code, and it cannot rescue a vague task or a missing test. It keeps the process honest and visible; getting the spec right is still your job.

Free and open source under the MIT license. No account, no API key, no paid tier — it runs on the AI subscription you already have.