GitHub Action

The fleetspark/fleet-action GitHub Action lets you run fleet operations inside GitHub Actions workflows. GitHub-hosted runners become your ships — no extra machines required.

When to use it

• You want to try Fleet without setting up multiple physical machines
• Your CI already runs on GitHub Actions and you want fleet runs triggered by PRs or schedules
• You want to automate a recurring fleet task (weekly dependency updates, nightly test coverage runs)

Installation

No installation needed. Reference the action directly in your workflow:

uses: fleetspark/fleet-action@v1

Basic example

.github/workflows/fleet.yml

name: Fleet Run

on:
  workflow_dispatch:
    inputs:
      plan:
        description: 'Goal to decompose into missions'
        required: false
        default: ''
      template:
        description: 'Built-in template name (test-coverage, security-audit, api-docs, dependency-update, refactor, drsti-dev-flow)'
        required: false
        default: ''

jobs:
  # Step 1: Commander plans the missions
  fleet-plan:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: fleetspark/fleet-action@v1
        with:
          command: plan
          plan: ${{ github.event.inputs.plan }}
          template: ${{ github.event.inputs.template }}
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}

  # Step 2: Ships execute in parallel
  fleet-ship:
    needs: fleet-plan
    runs-on: ubuntu-latest
    strategy:
      matrix:
        ship: [1, 2, 3]
    steps:
      - uses: actions/checkout@v4
      - uses: fleetspark/fleet-action@v1
        with:
          command: ship
          agent: claude-code
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}

Template-based run (no API key for planning)

Using a built-in template skips LLM decomposition — no ANTHROPIC_API_KEY needed for the plan step:

- uses: fleetspark/fleet-action@v1
  with:
    command: plan
    template: test-coverage

Inputs

Input	Description	Required
command	Operation to run: init, plan, ship, status, report	Yes
plan	Natural language goal (used with command: plan)	No
template	Built-in template name (used with command: plan)	No
agent	Coding agent to use (used with command: ship): claude-code, codex, aider, opencode, gemini, cursor, amp	No
repo	Git URL of the repository (defaults to current repo)	No

Generating a run report

Add a report step at the end of your workflow:

  fleet-report:
    needs: fleet-ship
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: fleetspark/fleet-action@v1
        with:
          command: report
      - uses: actions/upload-artifact@v4
        with:
          name: fleet-report
          path: fleet-report.md

Secrets

Secret	When needed
ANTHROPIC_API_KEY	Required when using --plan with LLM decomposition
GH_TOKEN	Required for the action to create PRs (auto-provided by Actions as GITHUB_TOKEN)

Limitations

• GitHub-hosted runners are ephemeral — each ship starts from a fresh checkout
• No persistent state between runs beyond what’s committed to the fleet/state branch
• Free-tier GitHub Actions concurrency limits apply to the matrix strategy
• Agent setup (Claude Code, Codex, etc.) must be handled in your workflow steps before the ship command

Source

The action source is in the action/ directory at the root of the FleetSpark repository.