Agent CI is a local GitHub Actions runner built for AI-agent development workflows. The current product focuses on running workflows on your machine with near-instant cache reuse, pausing on failures so an agent can fix the issue in place, and retrying without pushing commits to trigger remote CI.
Features
- Local execution of GitHub Actions workflows with bind-mounted caches for fast reruns
- Pause-on-failure flow that preserves container state so fixes can be applied before retrying
- Retry support for failed steps instead of restarting the whole workflow from scratch
- Emulation layer around the official GitHub Actions runner binary rather than a full custom reimplementation
- Guidance for integrating local CI runs into agent instruction files like
AGENTS.mdorCLAUDE.md
So agents can…
- Run CI checks locally before pushing code and cut feedback loops from minutes to seconds
- Inspect and fix failing workflow steps without losing the environment state
- Treat GitHub Actions as part of an iterative local agent workflow instead of a remote-only gate
Comments