Run AI agent on CI failures to fix them #831

We could trigger an action when scheduled checks fail, to run an agent that will fix the CI automatically (in a draft pull request). It's the second time that CI fails and it's trivial for an AI to fix the issue, but it took us ~24h to realize the problem, start an agent and merge the result. It's also a step in the direction of automating the maintenance of this project (which increases chances of it being maintained for years, reduces our workload, and I'm also curious about the extent to which this is possible).

The agent needs internet access, read access to the commit history, to AGENTS.md, and so on. I'm not sure what our current setup of opencode has access to right now. It should also have the possibility to re-run the CI in case the bug is just some temporary resource unavailability.