anthropics/claude-code-action

This repository is a GitHub Action that runs Claude Code inside a repository's own GitHub runner. It watches the workflow context and decides whether to act on an @claude mention, an assigned issue, or an explicit automation prompt, then can answer questions, review a PR, or commit code changes. It is aimed at teams that already work through GitHub pull requests and issues and want an assistant to participate in that same flow rather than in a separate chat window.

What changes in practice

Once the action is installed in a workflow, a comment mentioning @claude becomes a task: Claude can answer questions about the code or architecture, review changes, or attempt fixes, refactors, and small features README. Because the action reasons about workflow context, it selects its execution mode automatically instead of needing separate configurations for the mention, assignment, and automation cases README. Progress checkboxes are posted in the thread and update as the task runs, so reviewers can follow along without polling.

For repeated automation, the action also supports explicit prompts and structured JSON outputs that become GitHub Action outputs, which lets a later workflow step consume the result. The solutions guide documents patterns such as automatic PR review, path-specific review triggers, scheduled maintenance, issue triage and labeling, documentation sync, and security-focused review Solutions guide.

Setup and provider choices

The quickest path starts from Claude Code in the terminal with /install-github-app, which walks through installing the GitHub app and adding the required secrets. That shortcut is limited to direct Anthropic API users and requires repository admin rights; Bedrock, Vertex AI, and Microsoft Foundry users must follow the separate cloud-provider instructions README, Cloud providers. Authentication options include an API key, workload identity federation, or one of those cloud providers README. The action itself executes on the GitHub runner, while model calls go to the chosen provider.

Teams upgrading an existing v0.x workflow should read the migration guide before switching, since v1.0 simplifies the prompt and claude_args inputs and the guide describes the step-by-step workflow changes Migration guide.

Conditions to check

Access control and commit signing are treated as their own topic in the security documentation, and the capabilities and limitations page states what Claude can and cannot do in this setting Security, Capabilities & limitations. Note that the README does not itself summarize those limits, so they should be read before deciding how much autonomy to grant. The README also points to a configuration guide for MCP servers, permissions, and environment variables README. The project is MIT-licensed. Star counts shown for this repository are dated snapshots from a GitHub trending listing, not a measure of reliability or readiness.