CI-Failure Repair Loop

What your AI can do with it

• Wakes an agent when a build fails, passing the failure logs as its contract
• Reproduces the CI failure before attempting a fix
• Fixes the failure on a branch rather than directly on the main branch
• Pushes the fix automatically once the agent has it working
• Bounds the repair loop so the agent cannot thrash indefinitely

Getting started

1. Have a CI setup whose failure logs can be handed to the automation.
2. Add the CI-Failure Repair Loop to your agent's available tools.
3. Configure the agent so it receives the failure logs as its contract when a build goes red.
4. Set the bounds on the loop so it stops after a limited number of attempts.
5. Let a build fail and confirm the agent reproduces, fixes, and pushes on a branch.

Signals

GitHub stars

55k

Forks

5k

Last commit

Sep 2026

Questions

Does the agent push directly to the main branch?

No. The agent fixes the failure on a branch and pushes there, so the fix goes through normal review.

What stops the agent from looping forever on a hard failure?

The loop is bounded, so the agent cannot thrash indefinitely on a build it cannot fix.

What does the agent work from when a build fails?

It is woken with the failure logs as its contract, and it must reproduce the failure before fixing it.