claude-code-github-action

---
name: claude-code-github-action
description: Use when running Claude inside GitHub Actions — wires @claude PR review, issue-to-PR loops, scheduled audits, and auto-fixes with the official anthropics/claude-code-action.
source: https://github.com/anthropics/claude-code-action
generated: 2026-05-17T04:18:14.610Z
category: tool
audience: devops
---

## When to use

- Adding automated PR review on opened/synchronize events
- Letting collaborators trigger Claude with @claude mentions in issues or PRs
- Running a scheduled repo health check on a cron
- Wiring Claude with AWS Bedrock, Google Vertex AI, or Microsoft Foundry instead of the Anthropic API
- Limiting per-run cost via budget_tokens and timeout
- Upgrading an existing v0.x workflow to v1

## Key concepts

### Intelligent mode detection

The action picks its execution mode automatically based on workflow context — @claude mention vs PR opened vs issue assigned vs scheduled run — so a single uses: anthropics/claude-code-action@v1 step covers review, Q&A, and implementation without per-mode wiring.

### Runs on your runner

Execution happens entirely on your GitHub Actions runner. The action checks out the repo, talks to Claude, edits files, and posts comments using your GITHUB_TOKEN — no hosted Anthropic worker touches your code.

### Multi-provider auth

Supports four auth paths: direct Anthropic API (ANTHROPIC_API_KEY), AWS Bedrock (AWS_ROLE_ARN + AWS_REGION via OIDC), Google Vertex AI (GOOGLE_CLOUD_PROJECT), and Microsoft Foundry (AZURE_TENANT_ID). Cloud-provider auth uses GitHub OIDC, which is why id-token: write is required.

### claude_args JSON

Optional JSON blob passed as a workflow input to tune the Claude call: budget_tokens caps spend per run, timeout caps wall time in seconds, temperature controls determinism. Used to keep CI runs predictable and cheap.

### Structured outputs

The action emits validated JSON results as GitHub Action outputs, so downstream steps can branch on Claude's findings (e.g., post a check, open a PR, fail the build) instead of scraping comment text.

### Progress checkboxes

Long-running tasks render a live progress comment with dynamic checkboxes, so PR reviewers can watch Claude work through the plan rather than waiting for a single final blob.

## API reference

```
uses: anthropics/claude-code-action@v1
```

Run Claude inside a workflow job. Requires a prompt input plus ANTHROPIC_API_KEY (or a cloud-provider auth set), and the right permissions block.

```
name: Auto PR Review
on:
  pull_request:
    types: [opened, synchronize]

jobs:
  review:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write
      id-token: write
    steps:
      - uses: actions/checkout@v4
      - uses: anthropics/claude-code-action@v1
        with:
          prompt: "Review this PR for code quality and security"
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
```

```
on: issue_comment + if: contains(... '@claude')
```

Fire Claude only when a comment mentions @claude. Standard pattern for interactive issue/PR assistants.

```
on:
  issue_comment:
    types: [created]

jobs:
  claude:
    if: contains(github.event.comment.body, '@claude')
    runs-on: ubuntu-latest
    permissions:
      issues: write
      contents: read
      id-token: write
    steps:
      - uses: actions/checkout@v4
      - uses: anthropics/claude-code-action@v1
        with:
          prompt: ${{ github.event.comment.body }}
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
```

```
claude_args: '{ "budget_tokens": ..., "timeout": ..., "temperature": ... }'
```

Cap cost and runtime per invocation. budget_tokens is the hard token ceiling, timeout is wall-clock seconds, temperature tunes determinism.

```
- uses: anthropics/claude-code-action@v1
  with:
    prompt: "Refactor this code for better performance"
    claude_args: |
      {
        "budget_tokens": 50000,
        "timeout": 600,
        "temperature": 0.7
      }
```

```
claude /install-github-app
```

One-shot setup from the Claude Code terminal — installs the GitHub App and wires the workflow. Requires repo admin access.

```
claude /install-github-app
```

## Gotchas

- id-token: write is required for any OIDC-based auth (Bedrock, Vertex AI, Foundry) — omitting it silently fails auth
- pull-requests: write and issues: write are required for Claude to post comments; without them you get a 'Permission denied' on the first comment attempt
- Avoid contents: write at the repo level — scope it to the job that actually needs to push, the rest should stay contents: read
- Large repos are slow to checkout with fetch-depth: 0 — only use full history when Claude truly needs it
- API rate limits bite on busy repos — use budget_tokens and exponential backoff to avoid noisy failures
- Commits authored by the action are not GPG-signed by default; set up SIGNING_KEY if your branch protection requires signed commits
- v0.x to v1 is a breaking upgrade — follow docs/migration-guide.md instead of bumping the tag blindly

---
Generated by SkillMake from https://github.com/anthropics/claude-code-action on 2026-05-17T04:18:14.610Z.
Verify against source before relying on details.

File: ~/.claude/skills/claude-code-github-action/SKILL.md