Instead of using external issue trackers, tickets live directly inside the repository as Markdown files (/tickets/T-000123.md). State changes are commits. Validation runs in pre-commit and CI.

The idea is simple:

One ticket = one file

Strict lifecycle: ready → claimed → needs_review → done

Dependency-aware task picking

WIP limits enforced

mt pick lets agents self-select the best ready task

Mermaid graph output for dependencies

JSON export for LLM tooling

It’s optimized for multi-agent workflows where coordination becomes the bottleneck rather than code generation.

There’s no database, no SaaS backend, and no sync layer. Git becomes the coordination substrate.

install the repo as submodule and ask your AI agent to implement it