GitHub Actions Setup for Claude Code Integration (Archived)

📌 Important: This document has been archived. The Gemini CI workflow has been disabled, and self-hosted runners are currently not in use.

Current Status:

• ✅ Migrated to GitHub-hosted runners
• ✅ Gemini CI workflow disabled (saved as .disabled file)
• ✅ All CI/CD features continue to operate on GitHub-hosted runners

New Document:

• English version: docs/github-actions-setup-en.md
• Japanese version (current archive info): docs/gemini-cli-setup.md

---

Previous Information: This document described the procedure to set up Gemini CLI to work using claude-code-router when a Claude Code Action is executed on a self-hosted runner during issue creation.

Overview

Issue Analysis Feature

• Workflow: .github/workflows/issue-claude-action.yml
• Trigger: Upon Issue creation/edit, or comment creation containing @claude
• Execution Environment: Self-hosted runner
• AI Model: Gemini 2.5 Pro (via Gemini CLI)
• Routing: Uses Claude Code Router

PR Automated Test Fix Feature

• Workflow: .github/workflows/pr-test-fix.yml
• Trigger: When a PR test fails
• Execution Environment: Self-hosted runner
• AI Model: Gemini 2.5 Pro (via Gemini CLI)
• Feature: Automatically analyzes and fixes test failures, repeating until tests pass

Prerequisites

1. Self-hosted Runner Setup

You need to add a self-hosted runner to your GitHub repository:

1. Go to Repository Settings > Actions > Runners
2. Click "New self-hosted runner"
3. Follow the instructions to configure the runner

2. Runner Environment Preparation

Prepare the following environment on the self-hosted runner:

# Install Node.js 22+
curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -
sudo apt-get install -y nodejs

# Install required global packages
npm install -g @google/gemini-cli
npm install -g @musistudio/claude-code-router

# Git configuration (for GitHub Actions)
git config --global user.email "github-actions[bot]@users.noreply.github.com"
git config --global user.name "GitHub Actions Bot"

3. Gemini CLI Authentication Setup

Method 1: OAuth Authentication (Recommended)

Run the following on the runner:

gemini auth login

A browser will open; login with your Google account and complete the authentication. Authentication credentials will be saved in ~/.gemini/oauth_creds.json.

Method 2: API Key Authentication

1. Get an API Key at Google AI Studio
2. Go to GitHub repository Settings > Secrets and variables > Actions
3. Add the API Key with the name GEMINI_API_KEY

Workflow Behavior

Issue Analysis Workflow

Trigger Conditions

The workflow is executed when:

• An Issue is created or edited
• An Issue comment is created and contains @claude

PR Automated Test Fix Workflow

Trigger Conditions

The workflow is executed when:

• A PR test workflow completes with failure
• The target PR is open

Execution Steps

1. Environment Prep: Node.js, Git settings
2. Claude Code Router Setup: Create config file, retrieve transformer
3. Gemini CLI Auth Check: Verify OAuth or API Key authentication
4. Start Router: Start Claude Code Router in the background
5. Run Claude Code Action: Analyze Issue and generate response
6. Cleanup: Terminate Router process

Configuration File

The workflow starts Claude Code Router with the following settings:

Important Settings:

• Custom Transformer: Uses a custom transformer to invoke the actual Gemini CLI command
• forceModel: true: Passes --force-model option to Gemini CLI to enforce model usage
• "forceModel": "gemini-2.5-pro": Enforces specific model at the Router level
• MCP Support: Supports advanced features like MCP by using the actual Gemini CLI

{
    "LOG": true,
    "API_TIMEOUT_MS": 600000,
    "transformers": [
        {
            "path": "$HOME/.claude-code-router/plugins/gemini-cli-direct.js",
            "options": {
                "project": "outliner-d57b0",
                "forceModel": true
            }
        }
    ],
    "Providers": [
        {
            "name": "gemini-cli-direct",
            "api_base_url": "http://localhost:3456",
            "api_key": "dummy-key",
            "models": ["gemini-2.5-flash", "gemini-2.5-pro"],
            "transformer": {
                "use": ["gemini-cli-direct"]
            }
        }
    ],
    "Router": {
        "default": "gemini-cli-direct,gemini-2.5-pro",
        "think": "gemini-cli-direct,gemini-2.5-pro",
        "longContext": "gemini-cli-direct,gemini-2.5-pro",
        "longContextThreshold": 60000,
        "forceModel": "gemini-2.5-pro"
    }
}

Custom Gemini CLI Transformer

This project uses a custom transformer that calls the actual Gemini CLI command:

Features:

• Actual Gemini CLI Execution: Executes gemini command instead of Google Cloud Code API
• MCP Support: Supports full capabilities of Gemini CLI (MCP etc.)
• Force-Model Support: Enforces model usage with --force-model option
• JSON Output: Retrieves structured output with --format json
• Temp File Management: Manages conversation data in temporary files and automatically deletes them after execution

Execution Example:

gemini --model gemini-2.5-pro --force-model --project outliner-d57b0 --format json --file /tmp/conversation.json

Troubleshooting

Authentication Error

❌ No Gemini CLI credentials found

Solution:

1. Run gemini auth login on the Runner
2. Or set the GEMINI_API_KEY secret

Router Start Error

❌ Claude Code Router failed to start within 30 seconds

Solution:

1. Manually run ccr start on the Runner to check for errors
2. Check if port 3456 is available
3. Check the log file ~/.claude-code-router.log

Network Error

Solution:

1. Check if access to Google API is possible from the Runner environment
2. Add proxy settings to the configuration file if necessary

Security Considerations

• Execute self-hosted runners in a trusted environment
• Manage API Keys appropriately as secrets
• Protect OAuth credentials with appropriate file permissions
• Ensure log files do not contain sensitive information

Usage

1. Claude Code Action runs automatically when a new Issue is created
2. Analysis runs when a comment containing @claude is added to an existing Issue
3. Analysis results are added as Issue comments

Limitations

• Works only on self-hosted runners
• Requires Gemini CLI authentication
• Requires internet connection
• Concurrent execution is limited to one per Issue