Claude Composite Actions¶

Traditional GitHub Actions wrapping Claude Code. Consumed via uses: in standard workflows.

Note: This directory was renamed from workflows/ to claude-workflows/. A symlink ensures backwards compatibility — existing uses: elastic/ai-github-actions/claude-workflows/...@v0 references continue to work.

Security¶

Important: RWX and RWXP agents can run arbitrary commands, including git commit and git push. To prevent pushes, set contents: read (or contents: none) in your workflow's permissions:. Prompt constraints alone are not sufficient.

Workflows triggered by user actions should restrict who can trigger Claude. The example workflows include an author_association check allowing only OWNER, MEMBER, and COLLABORATOR. See SECURITY.md for detailed guidance.

Available actions¶

Review agents¶

Action	Description	R	W	X	Git	Push
Issue Triage (RO)	Triage new issues (read-only)	✅	❌	❌	✅	❌
Issue Triage (RWX)	Triage new issues (can execute tests)	✅	✅	✅	✅	❌**
PR Review (RO)	Review PRs (suggestions only)	✅	❌	❌	✅	❌
PR Review (RWX)	Review PRs (with test execution)	✅	✅	✅	✅	❌**

Assistant agents¶

Action	Description	R	W	X	Git	Push
Build Failure Buildkite (RWX)	Analyze Buildkite CI failures	✅	✅	⚙️	✅	❌
Build Failure GitHub Actions (RWX)	Analyze GitHub Actions failures	✅	✅	⚙️	✅	❌
Mention in Issue (RWX)	@claude in issues (no push)	✅	✅	✅	✅	❌**
Mention in Issue (RWXP)	@claude in issues (full access)	✅	✅	✅	✅	✅
Mention in PR (RWX)	@claude in PRs (no push)	✅	✅	✅	✅	❌**
Mention in PR (RWXP)	@claude in PRs (full access)	✅	✅	✅	✅	✅

Overview agents¶

Action	Description	R	W	X	Git	Push
Generate Report (RO)	Scheduled report generation	✅	❌	⚙️	✅	❌
Project Manager (RO)	Periodic project state reviews	✅	❌	❌	✅	❌

✅ = default, ⚙️ = via config, ❌ = unavailable, ❌** = prompt-discouraged (enforce with contents: read)

Usage¶

- uses: elastic/ai-github-actions/claude-workflows/pr-review/rwx@v0
  with:
    prompt: "Review this PR"
    claude-oauth-token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
    github-token: ${{ github.token }}

Base action¶

The base action provides full configurability:

Input	Description	Required	Default
`prompt`	Prompt to pass to Claude	Yes	—
`claude-oauth-token`	Claude OAuth token	Yes	—
`github-token`	GitHub token for Claude	Yes	—
`allowed-tools`	Comma-separated allowed tools	No	`""`
`model`	Model to use	No	`claude-opus-4-5-20251101`
`allowed-bots`	Allowed bot usernames, or `*` for all	No	`""`
`mcp-servers`	MCP server configuration JSON	No	See below

MCP servers¶

All workflow actions include default MCP servers (the base action does not):

agents-md-generator — Generates repository summaries from AGENTS.md. Called automatically at startup.
public-code-search — Search public GitHub repos for usage patterns and reference implementations.

Override with your own:

- uses: elastic/ai-github-actions/claude-workflows/base@v0
  with:
    prompt: "Your prompt"
    claude-oauth-token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
    mcp-servers: '{"mcpServers":{"my-server":{"type":"http","url":"https://my-server.example.com/mcp"}}}'