31 languages, plus SQL lineage across Snowflake, dbt, and T-SQL. Local, free, MIT, never uploaded.
Anthropic's workshop on building agents that run for hours. The loop it describes (find the work, hand it to the agent, check the result, record state, decide the next move) is the loop this config is built around. The pieces it names map straight onto what installs:
| The loop needs | Atomic Claude ships |
|---|---|
| An automation that drives the work | /autopilot runs plan → implement → review → ship hands-off; ship verbs refresh signals on every commit |
| A skill that carries project context | Signals: a standing repo model Claude reads before your code, kept fresh automatically |
| A maker and a separate checker | atomic-implementer writes; atomic-reviewer re-runs the tests and gates the diff. The author never grades its own homework |
| A state file that survives the session | signals.md, the scratchpad STATE.md, and committed follow-ups hold what's done and what's next |
| An objective gate, not an opinion | atomic-tdd (failing test first) and atomic-verify (no "done" without a fresh run) |
| Worktrees for parallel work without collisions | The implement loop isolates each branch under .worktrees/ |
The video explains the pattern. This config is that pattern.
You have seen what it is and the pattern it runs. The next step is to feel it on your own repo: turn on the output style, index your code, and hand the loop a real task, from install to a first merged change.
