Skip to main content
Recently autodev.yml is renamed to devbird.yml because the product is rebranded. But for backward compatibility, autodev.yml will just work as before.

​
Prerequisites

Before you begin, ensure you have:
  • Completed the GitHub App installation
  • Admin or write access to your repository
  • Required secrets configured in your GitHub repository

​
Setting up the workflow

Follow these steps to add the DevBird workflow to your repository:

​
1. Create the workflow file

Create a new file at .github/workflows/devbird.yml in your repository.

​
2. Add the workflow configuration

Copy the following YAML configuration into the file: 
name: "DevBird"
run-name: "DevBird: ${{ inputs.task_title }}"

on:
  workflow_dispatch:
    inputs:
      prompt:
        description: "Instructions for DevBird. Can be a direct prompt or custom template."
        type: string
        required: true

      base_branch:
        description: "The branch to use as the base/source when creating new branches (defaults to repository default branch)"
        type: string
        required: false
        default: "main"

      agent:
        description: "The agent to use for the action. Can be 'claude_code', 'gemini_cli', 'codex_cli' or 'opencode'"
        type: choice
        default: "claude_code"
        options:
          - claude_code
          - gemini_cli
          - codex_cli
          - opencode
          - crush_cli
          - github_copilot_cli

      agent_model:
        description: "The (optional) model to use for the agent"
        type: string
        required: false
        default: ""

      devbird_workflow_execution_token:
        description: "The token to use for the DevBird task"
        type: string
        required: false
        default: ""

      devbird_mode:
        description: "The DevBird execution mode. Can be 'develop' (default) or 'plan' (for task graph planning)"
        type: choice
        default: "develop"
        options:
          - develop
          - plan

      task_title:
        description: "The title of the DevBird task"
        type: string
        required: false
        default: ""

jobs:
  devbird:
    runs-on: ubuntu-latest
    env:
      DEVBIRD: 1
    permissions:
      contents: write
      pull-requests: write
      issues: write
      id-token: write
      actions: read # Required for Claude to read CI results on PRs
    steps:
      - name: Checkout code
        uses: actions/checkout@v5

      # Setup step.
      # See next section for the documentation
      # - name: Setup Node.js
      #   uses: actions/setup-node

      - name: Run DevBird
        uses: delinoio/devbird-action@main
        with:
          agent: ${{ inputs.agent }}
          agent_model: ${{ inputs.agent_model }}
          devbird_mode: ${{ inputs.devbird_mode }}
          devbird_workflow_execution_token: ${{ inputs.devbird_workflow_execution_token }}
          prompt: ${{ inputs.prompt }}
          base_branch: ${{ inputs.base_branch }}
          delino_access_token: ${{ secrets.DELINO_ACCESS_TOKEN }}
          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
          openai_api_key: ${{ secrets.OPENAI_API_KEY }}

​
3. Customize the setup steps

The workflow includes a setup step that you should customize for your repository: 
- name: Setup Node.js
  uses: actions/setup-node
Replace these with your project’s setup steps. The steps here should configure a complete development environment for the AI agent.
For example, the Delino monorepo includes steps for
  • installing protoc for RPC definitions
  • installing golang for backend servers
  • installing nodejs for next.js apps
  • fetching environment variables for development environment

​
4. Filter environment variables for security

Security Best PracticePassing all environment variables to AI agents is not recommended. You should filter environment variables to only include those that are safe to expose.
Delino’s official repositories use a filtering approach to ensure only safe environment variables are available during DevBird execution. Add a filtering step in your setup to restrict which environment variables the AI agent can access: 
# Filter environment variables in DevBird mode
if [ -n "${DEVBIRD:-}" ]; then
    echo "DevBird mode detected - filtering environment variables to NEXT_PUBLIC_* and TEST_* only"
    if [ -f ".env" ]; then
        grep -E '^(NEXT_PUBLIC_|TEST_)' .env > .env.filtered || true
        mv .env.filtered .env
    fi
fi
This script filters the .env file to only include variables with these prefixes:
  • NEXT_PUBLIC_: Variables that are meant to be publicly exposed (e.g., for Next.js client-side code)
  • TEST_: Variables used for E2E testing, which have minimal security risk if leaked
The DEVBIRD environment variable (set to 1 in the workflow configuration) triggers this filtering behavior, ensuring your sensitive credentials remain protected.
You can customize the filtering pattern to match your project’s naming conventions. The key is to only expose environment variables that are safe for the AI agent to access.

​
5. Configure DELINO_ACCESS_TOKEN

Create a new API key following the guide at Creating API Keys and add it as a repository secret on GitHub settings.

​
6. Commit to your main branch

Commit the workflow file to your repository’s main branch: 
git add .github/workflows/devbird.yml
git commit -m "Add DevBird workflow"
git push

​
Understanding workflow execution

Important: Workflow runs on the main branchFor security reasons, the DevBird GitHub Action always executes on your repository’s default branch (typically main). This means:
  • The workflow runs from the main branch codebase
  • GitHub Actions logs will show the workflow running on main
  • This is by design to prevent security vulnerabilities
The base_branch parameter controls which branch DevBird uses as the base when creating new feature branches, but the workflow execution itself always happens on the default branch.
This security design prevents potential issues such as:
  • Malicious workflow modifications in feature branches
  • Unauthorized access to repository secrets
  • Tampering with the DevBird workflow configuration

​
Next steps

Once the workflow is set up, you can start using DevBird to automate development tasks. Learn how to trigger your first DevBird task in the Creating your first PR guide.