I built a GitHub Action that auto-fixes CI failures using Claude.
CI fails on a PR → Claude analyzes the error → fixes it → commits.
No context switch. No human in the loop for mundane stuff.
The best AI workflow is the one you never have to trigger.
🔗 gist.github.com/ArnaudRinqui...
![## Memory Suggestions
- When you notice repeated corrections, preferences, or patterns in my
feedback, proactively suggest: "Should I remember: [concise rule]?"
- Suggest memories for: code style preferences, architectural decisions,
workflow preferences, tools I prefer, things I consistently reject
- Format suggestions as single-line rules ready to paste into CLAUDE.md
- Batch multiple suggestions at natural stopping points (end of task, after
several corrections)](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:vxizjecldw4ka6lk5g46eunj/bafkreih55ekgwtfgjpdcn3ikr24jtpnozp3laxpcuvrtq4kqf4row6jfce)
