Vasishtha15/self-healing-cicd-agent — reverse-engineered prompt
Reverse engineered prompt
Build me a GitHub Actions based self healing CI helper for pull requests. I want it to run when a PR to main has a pipeline failure, collect the logs from install, lint, tests, and Docker build steps, figure out what kind of failure happened, send the useful log context to Claude, and then post a clear PR comment with the root cause, confidence, severity, suggested fixes, and a short prevention tip.
Keep it in Python and include a small sample app so the workflow can be tested with intentional failures. It should handle common dependency, lint, test, and Docker errors, and the PR comment should feel structured and actionable, not just dump raw logs. Use the repo secret for the Anthropic API key and the normal GitHub token for posting comments.
Please wire up the workflow so the analysis still runs even if the earlier CI steps fail, and make it easy to extend later with more failure types or other notifications. If you need details, look up current GitHub Actions and Anthropic docs online.
Want more depth? Deep Reverse