dobbyphus

A reusable GitHub Action that runs AI agents using opencode + oh-my-opencode.

Quick Start

name: AI Agent

on:
  workflow_dispatch:
    inputs:
      prompt:
        description: Custom prompt to run
        required: false
      issue_number:
        description: Issue/PR number to comment on
        required: false

  issue_comment:
    types: [created]

  pull_request:
    types: [review_requested]

  pull_request_review:
    types: [submitted]

  pull_request_review_comment:
    types: [created]

jobs:
  agent:
    if: >-
      github.event_name == 'workflow_dispatch' ||
      (github.event_name == 'issue_comment' &&
       contains(github.event.comment.body, '@ai-agent') &&
       github.event.comment.user.login != 'github-actions[bot]' &&
       contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association)) ||
      (github.event_name == 'pull_request' &&
       github.event.requested_reviewer.login == 'github-actions[bot]') ||
      (github.event_name == 'pull_request_review' &&
       github.event.review.user.login != 'github-actions[bot]' &&
       contains(github.event.review.body, '@ai-agent')) ||
      (github.event_name == 'pull_request_review_comment' &&
       contains(github.event.comment.body, '@ai-agent') &&
       github.event.comment.user.login != 'github-actions[bot]' &&
       contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association))
    runs-on: ubuntu-latest
    permissions:
      contents: write
      issues: write
      pull-requests: write

    steps:
      - uses: actions/checkout@v6
        with:
          ref: ${{ github.head_ref || github.ref }}
          fetch-depth: 0

      - uses: dobbyphus/action@main
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          bot_name: ai-agent

See examples/agent.yaml for a complete workflow with mode detection and GitHub App token comments.

Inputs

Input	Default	Description
bot_name	ai-agent	Bot name for labels and mentions
mention_users	false	Whether to @mention users in comments (reduces notification noise when false)
anthropic_api_key	-	Anthropic API key
openai_api_key	-	OpenAI API key
gemini_api_key	-	Google Gemini API key
provider_anthropic	max20	Anthropic provider setting (no/yes/max20)
provider_openai	no	OpenAI provider setting (no/yes)
provider_google	no	Google provider setting (no/yes)
provider_copilot	no	GitHub Copilot provider setting (no/yes)
anthropic_base_url	-	Custom Anthropic API base URL (for proxies)
openai_base_url	-	Custom OpenAI API base URL (for proxies)
primary_model	-	Override opencode.json model
mode	agent	Prompt mode (maps to {prompt_path}/{mode}.md)
prompt	-	Direct prompt text (alternative to mode)
prompt_path	.github/prompts	Path to prompts directory
prompt_vars	-	JSON object for template substitution
github_token	github.token	GitHub token for API access
opencode_version	latest	OpenCode version to install
oh_my_opencode_version	latest	oh-my-opencode version to install
config_json	-	Full opencode.json content (advanced)
enabled_providers	-	JSON array of provider IDs to enable
disabled_providers	-	JSON array of provider IDs to disable
omo_config_json	-	Full oh-my-opencode.json content (advanced)
auth_json	-	Full auth.json content (advanced)
agent_keywords	ultrawork	Keywords to prepend for agent mode (triggers oh-my-opencode modes)
review_keywords	analyze	Keywords to prepend for review mode (triggers oh-my-opencode modes)
skill_enable_git_master	false	Enable git-master builtin skill (true/false)
skill_enable_playwright	false	Enable playwright builtin skill (true/false)
skill_enable_frontend_ui_ux	false	Enable frontend-ui-ux builtin skill (true/false)
format_output	true	Format output with collapsible sections for GitHub Actions logs

How It Works

The action handles everything in a single step:

1. Setup: Collects context from the trigger event, configures git, adds 👀 reaction
2. Run: Executes the AI agent with the configured prompt
3. Teardown: Replays commits as signed, creates PRs, updates reaction to 👍 or 😕

Modes

The action supports two modes via the mode input:

Mode	Use Case	Prompt
agent	Issue comments, PR comments, workflow dispatch	Work on requests, make changes
review	PR opened, review requested	Review code, provide feedback

Trigger Conditions

Use job-level if to control when the agent runs. The example covers:

• workflow_dispatch: Always run on manual trigger
• issue_comment: @mention by authorized user
• pull_request: Non-draft PR opened or ready for review
• pull_request_review: @mention in review body
• pull_request_review_comment: @mention in inline diff comment

Branch & PR Workflow

The agent must create a branch for any changes. Direct commits to the default branch are blocked.

1. Agent creates a branch: git checkout -b fix/issue-42
2. Agent makes commits with meaningful messages
3. Action automatically:
   • Replays commits as signed via GitHub API
   • Creates the branch on remote if new
   • Opens a pull request to the default branch

Signed Commits

All commits are signed and verified by GitHub. The agent commits normally using git commit, and the action replays each commit through the GitHub API, which signs them automatically.

Workflow by Trigger

Trigger	Mode	Agent Should	Action Does
Issue comment	agent	Create branch, commit	Push branch, create PR
PR comment	agent	Commit to PR branch	Push to PR branch
PR inline comment	review	Review code, respond	Commit to PR branch
Workflow dispatch	agent	Create branch, commit	Push branch, create PR
PR opened	review	Review code	Post review comment
Review request	review	Review code	Post review comment

Configuration

Repository Variables

Set these in Settings → Secrets and variables → Variables:

Variable	Default	Description
BOT_NAME	ai-agent	Bot mention trigger and label prefix
BOT_LOGIN	github-actions[bot]	Bot login to prevent self-triggering

Authentication

The action accepts a github_token input. Choose the right token for your use case:

Default Token

The simplest option. Uses the automatic GITHUB_TOKEN:

- uses: dobbyphus/action@main
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}

Limitations: Cannot trigger other workflows, limited API rate, commits attributed to github-actions[bot].

GitHub App Token

For production use. Provides higher API limits, can trigger workflows, and commits are attributed to your app:

- uses: actions/create-github-app-token@v1
  id: app-token
  with:
    app-id: ${{ vars.APP_ID }}
    private-key: ${{ secrets.APP_PRIVATE_KEY }}

- uses: actions/checkout@v6
  with:
    ref: ${{ github.head_ref || github.ref }}
    fetch-depth: 0
    token: ${{ steps.app-token.outputs.token }}

- uses: dobbyphus/action@main
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    github_token: ${{ steps.app-token.outputs.token }}

Setup: Create a GitHub App with contents: write, issues: write, pull-requests: write permissions. Store the App ID in a variable and the private key as a secret.

See the commented section in examples/agent.yaml for the complete pattern.

Personal Access Token

For individual use or cross-repository access:

- uses: dobbyphus/action@main
  with:
    anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
    github_token: ${{ secrets.PAT_TOKEN }}

Prompt Templating

Templates use {{ variable }} syntax with multi-pass resolution:

# {{ title }}

{{ base_github_env }}

Hello @{{ author }}, working on {{ task }}...

Base Snippets

Place .md files in prompts/base/ to create reusable snippets:

.github/prompts/
├── base/
│   └── my_rules.md   → {{ base_my_rules }}
├── agent.md
└── review.md

For project conventions, use AGENTS.md in the repository root instead. The agent checks for AGENTS.md, CLAUDE.md, and COPILOT.md automatically.

Override Priority

Variables are merged with later sources overriding earlier ones:

1. Action's prompts/base/*.md (lowest priority)
2. Consumer's prompt_path/base/*.md
3. User-provided prompt_vars (highest priority)

Development

python -m pytest tests/ -v

License

MIT