ReviewQueue

The problem

Coding agents can run shell commands autonomously — including git push, deploys, and destructive operations. One misread instruction can ship broken code or wipe data. You need a gate between the agent and the shell, with a clear audit trail.

How it works

1. 1. Agent proposes a commandRisky argv lists are classified and held in a local pending queue instead of running immediately.
2. 2. You review contextSee the command, working directory, git branch, and diff stat before deciding.
3. 3. Approve or rejectApproved commands execute once; every transition is logged to JSONL for compliance.

Pricing