CTX

CTX Fit analyzes a code repository, runs promising AI coding-agent configurations against real tasks in it, and produces the winning setup as a reviewable change. You can apply it directly to your working tree or open it as a pull request. It is aimed at developers and teams who use coding agents such as Claude Code and want evidence about which setup is cheapest while still working reliably, rather than choosing by guesswork.

The winner is picked by a fixed rule rather than a score: candidates below a reliability floor are discarded, the remaining ones are ranked by attributable cost, and ties go to the simpler configuration. If nothing beats your current setup, the tool says so. A language model may explain a result but never decides it. The scope is comparing capability configurations within one coding-agent harness, not comparing different agents with each other.

Verification uses repository-native commands for Python, JavaScript and TypeScript, Go, Rust and Make, and treats the chosen test command as the authority. Final verification runs in an isolated home without network access, so dependencies must already be available. The README is explicit that this is evidence for normal development and does not prove that hostile code cannot fool its own test runner. The project is written in Python and MIT licensed.