Agent CI offers a local, high-performance alternative to cloud-based GitHub Actions, specifically designed for developers and AI agents who require rapid feedback loops and efficient debugging. It tackles the inherent latency and cost of traditional CI by enabling workflows to execute entirely on the developer's machine. Key benefits include near-instantaneous caching, interactive debugging capabilities, and the ability to retry failed steps without re-running the entire job, dramatically accelerating the development and troubleshooting cycle.

How It Works

This project fundamentally replaces GitHub Actions' cloud orchestration layer with a local emulation of its server-side API, including Twirp endpoints and the Azure Block Blob protocol. This allows the unmodified, official GitHub Actions Runner binary to execute jobs locally, ensuring bit-for-bit compatibility. The core innovation lies in its caching mechanism, which utilizes local bind-mounts for dependencies (like node_modules, pnpm stores) and tool caches, eliminating network round-trips and achieving cache access times around ~0 ms. Furthermore, Agent CI introduces a unique "pause on failure" mode. Instead of tearing down the container, it preserves all state—environment variables, installed tools, and artifacts—allowing developers or AI agents to inspect the failure, make edits on the host synced into the container, and retry only the failed step.

Quick Start & Requirements

• Install: npm install -D @redwoodjs/agent-ci
• Prerequisites: Docker (OrbStack or Docker Desktop on macOS; Native Docker Engine on Linux).
• Run Workflow: npx agent-ci run --workflow .github/workflows/ci.yml or npx agent-ci run --all.
• Retry Failed Step: npx agent-ci retry --name <runner-name>
• Docs: CLI package documentation is available at packages/cli/README.md#docker-host-resolution-for-job-containers.

Highlighted Details

• ~0ms Caching: Achieved through direct local bind-mounts of host directories (e.g., node_modules, pnpm store, runner tool cache) into containers, bypassing upload/download overhead.
• Pause on Failure: Containers remain alive with full state preservation upon step failure, enabling interactive debugging and targeted retries of only the failed step.
• AI Agent Integration: Explicitly designed for AI-driven development loops, allowing agents to fix issues and retry locally without incurring remote CI costs.
• Working Tree Execution: Runs directly against the current working tree, automatically including uncommitted changes without requiring stashing or committing.

Maintenance & Community

No specific details regarding notable contributors, sponsorships, partnerships, or community channels (e.g., Discord, Slack) were found within the provided README.

Licensing & Compatibility

The README does not explicitly state a software license, which is a significant omission for assessing commercial use or integration compatibility. Compatibility is claimed to be high for any workflow that runs on GitHub.com, as it leverages the official runner binary and emulates the necessary API surface.

Limitations & Caveats

The most critical limitation is the absence of a clearly defined software license, making it difficult to ascertain usage rights, especially for commercial applications. The tool's operation is entirely dependent on a functional and correctly configured Docker environment.