Fix CI

Fix CI is one of the standard use cases of the Agentic SDLC, delivered as a playbook you import and adapt. It is an agentic workflow that starts from a single event, a build going red on a pull request, and runs the loop a developer would otherwise run by hand. Its diagnosis step is a root cause analysis over the CI logs: read what failed, work out why, and name the files involved before changing anything.

What separates it from a generic "self-healing pipeline" is where it draws the line. Fix CI assumes the pipeline is correct and fixes the code on the branch to pass it. It does not edit CI configuration, and when the real cause is the pipeline rather than the code, it stops and tells a human instead of papering over the problem. It also never merges its own work: the agent pushes a green branch and explains the change, and a person approves the merge. That is the Overcut pattern in miniature, real autonomy on the task inside firm guardrails, with a human at the gate.

An engineer opens a pull request that bumps a shared library to a new major version. CI goes red: three unit tests fail because a function signature changed. Instead of the author switching off their next task to investigate, the failed-build event triggers Fix CI. It clones the branch, reads the job logs, and sees the three failures trace to call sites that still pass the old arguments. It writes a fix plan, updates those call sites, runs the tests to confirm they pass, commits as fix(ci): update call sites for new signature, and pushes. It comments on the PR summarizing the change. The author comes back to a green build and a short diff, reviews it, and merges. They never left the work they were on, and nobody touched the CI configuration.