GitHub Actions

The anthropics/claude-code-action action runs Claude Code inside a GitHub Actions runner. Point it at CodeVector by setting ANTHROPIC_BASE_URL to your gateway’s /gateway/anthropic endpoint, then attribute the spend with custom headers. Every model call is billed against the API key’s user, the same as if it were running on a developer’s laptop.

Prerequisites

• A CodeVector instance reachable from GitHub-hosted runners (a public URL, or a runner inside your VPC).
• A per-seat API key with a Model grant for the slug you want the workflow to use. See API keys.
• A Model facade configured for code review or PR triage (for example, a slug like prreview-model that fans out to Claude Sonnet). See Models.

On-demand workflow: reply to @claude

This workflow runs when someone mentions @claude in an issue, PR comment, or review.

# .github/workflows/claude.yml
name: Claude Code

on:
  issue_comment:
    types: [created]
  pull_request_review_comment:
    types: [created]
  issues:
    types: [opened, assigned]
  pull_request_review:
    types: [submitted]

jobs:
  claude:
    if: |
      (github.event_name == 'issue_comment' && contains(github.event.comment.body, '@claude')) ||
      (github.event_name == 'pull_request_review_comment' && contains(github.event.comment.body, '@claude')) ||
      (github.event_name == 'pull_request_review' && contains(github.event.review.body, '@claude')) ||
      (github.event_name == 'issues' && (contains(github.event.issue.body, '@claude') || contains(github.event.issue.title, '@claude')))
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: read
      issues: read
      id-token: write
      actions: read

    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 1

      - uses: anthropics/claude-code-action@v1
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          settings: |
            {"env":{"ANTHROPIC_BASE_URL":"https://your-gateway.example.com/gateway/anthropic","ANTHROPIC_CUSTOM_HEADERS":"x-client-app: claude-code\nx-project: your-repo"},"model":"prreview-model"}
          additional_permissions: |
            actions: read

Replace your-gateway.example.com with your gateway hostname, your-repo with the repo name you want to attribute spend to, and prreview-model with a model slug that exists on your gateway.

Auto review every pull request

This workflow runs on every PR open, push, and reopen. It uses the code-review plugin to drive Claude through a structured review.

# .github/workflows/claude-code-review.yml
name: Claude Code Review

on:
  pull_request:
    types: [opened, synchronize, ready_for_review, reopened]

jobs:
  claude-review:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: read
      issues: read
      id-token: write

    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 1

      - uses: anthropics/claude-code-action@v1
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          settings: |
            {"env":{"ANTHROPIC_BASE_URL":"https://your-gateway.example.com/gateway/anthropic","ANTHROPIC_CUSTOM_HEADERS":"x-client-app: claude-code\nx-project: your-repo"},"model":"prreview-model"}
          plugin_marketplaces: "https://github.com/anthropics/claude-code.git"
          plugins: "code-review@claude-code-plugins"
          prompt: "/code-review:code-review ${{ github.repository }}/pull/${{ github.event.pull_request.number }}"

Filter by author, paths, or labels using the if: and paths: keys if you don’t want every PR reviewed. The action’s usage docs cover the available conditions.

The settings field

settings is a JSON string that claude-code-action parses and applies. The two fields that matter for gateway routing are env and model.

• env.ANTHROPIC_BASE_URL: the Anthropic surface of your gateway, ending in /gateway/anthropic. The action’s underlying SDK appends /v1/messages and other paths automatically.
• env.ANTHROPIC_CUSTOM_HEADERS: a newline-delimited string of Header: Value pairs. The Anthropic SDK splits on the literal newline that JSON encodes from \n. The gateway reads x-client-app and x-project; everything else passes through to the upstream provider unchanged.
• model: a Model facade slug on your gateway. Not the upstream Provider Model identifier. If the slug isn’t granted to the API key’s user, the request returns 403.

You can use a YAML block scalar to keep the line readable. The value must still be valid JSON:

settings: |
  {
    "env": {
      "ANTHROPIC_BASE_URL": "https://your-gateway.example.com/gateway/anthropic",
      "ANTHROPIC_CUSTOM_HEADERS": "x-client-app: claude-code\nx-project: your-repo"
    },
    "model": "prreview-model"
  }

Attribution headers

Every gateway request is recorded against the calling user, model, IDE, and project. The two headers above set the IDE and project dimensions:

• x-client-app: claude-code: identifies the calling tool. Use the same value across CI and developer laptops so admins can see total Claude Code spend in one slice.
• x-project: <repo-name>: identifies the codebase. Pick a stable slug per repo.

If your team needs PR-level attribution (e.g. spend per pull request), encode the PR number into x-project (x-project: your-repo/pr-123). This expands the cardinality of recorded usage; check with your admin before turning it on across the org.

Frequently asked questions

Which auth input should I use, anthropic_api_key or claude_code_oauth_token?

Use anthropic_api_key. The OAuth token input is for the Claude Code subscription and does not authenticate against your gateway. CodeVector expects a bearer API key, so pass your gateway key via anthropic_api_key.

Can I send the PR number, branch, or commit SHA as headers?

You can add them to ANTHROPIC_CUSTOM_HEADERS, but the gateway only reads x-client-app and x-project today. Other headers pass through to the upstream provider unchanged and are not recorded for attribution. Ask your admin if you need PR-level attribution.

How do I pick the right model slug?

Use any Model facade slug visible on the Available models page or returned by GET /v1/models on the gateway. modelId is the user-facing facade, not the upstream Provider Model. If the slug does not resolve, the workflow run fails with a 404 from the gateway.

Why is the settings field one giant JSON string?

claude-code-action accepts a JSON string for settings. The action parses it and sets the env block as process env vars. You can use a YAML block scalar to keep the line short, but the value must still be valid JSON.

What permissions does the workflow need?

contents: read, pull-requests: read, issues: read, and id-token: write. Add actions: read if you want Claude to read CI results on PRs. Never grant write permissions broader than what the workflow uses.

Related docs

• Connect. Wire Claude Code, opencode, and other tools on a developer laptop instead of CI.
• API keys. Create the per-seat key the workflow signs in with.
• Available models. The Model facade slugs you can pass to the action.
• Usage. See the spend each workflow run drives, sliced by clientApp and project.