Anvil

Anvil puts an AI coding agent to work on the system it runs on.

The design: a daily audit loop that reads the codebase, identifies technical debt and code quality issues, scores them by estimated impact and effort, and dispatches specialist subagents to fix them — without a human in the loop. Not a CI/CD pipeline, but a reasoning agent that makes editorial decisions about what to improve next and how.

Phase 0 is the audit loop: a single agent that reads, evaluates, and produces a prioritized work queue. Subsequent phases add the dispatcher (load-balancing work across parallel workers) and the specialist fleet (debugger, refiner, test-writer, documentation-architect). The aim is a system that gets measurably better at its own job over time, using Claude Code's agent architecture as the runtime.