Agents Docs
Back to Workflows

AI On-Demand Assistant

GitHub Actions workflow for AI-powered assistance triggered by comments

Author: EFP Team
github-actionsaiautomationclaude

AI On-Demand Assistant Workflow

A GitHub Actions workflow that provides AI-powered assistance triggered by mentioning a bot in issues, pull requests, and comments. This workflow integrates Claude Code with GitHub to provide contextual help to development teams.

Overview

The AI On-Demand Assistant workflow responds to mentions of @efp-dev-ops in:

  • Issue comments
  • Pull request review comments
  • Pull request reviews
  • New issues

When triggered, it extracts the user's request, processes it through Claude Code, and responds with helpful assistance.

Features

  • Multi-trigger Support: Responds to various GitHub events
  • User Authorization: Only processes requests from authorized users
  • Input Validation: Validates and sanitizes user input
  • Error Handling: Continues execution even if individual steps fail
  • Timeout Protection: Prevents long-running executions

Configuration

Required Secrets

  • APP_ID: GitHub App ID for token generation
  • PRIVATE_KEY: GitHub App private key
  • CLAUDE_CODE_OAUTH_TOKEN: OAuth token for Claude Code integration

Required Variables

  • ALLOWED_USER_LIST: JSON array of authorized GitHub usernames

Permissions

The workflow requires the following permissions:

  • contents: read - Read repository content
  • issues: write - Comment on issues
  • pull-requests: write - Comment on pull requests

Usage

  1. Setup: Configure the required secrets and variables in your repository
  2. Authorization: Add authorized users to the ALLOWED_USER_LIST variable
  3. Trigger: Mention @efp-dev-ops in an issue or pull request with your request

Example Usage

@efp-dev-ops Can you help me understand this code structure?
@efp-dev-ops Review this pull request and suggest improvements

Workflow Steps

1. Event Filtering

The workflow only runs when:

  • The mention @efp-dev-ops is present in the content
  • The triggering user is in the authorized user list
  • The request comes from a supported event type

2. Code Checkout

Checks out the repository code to provide context for AI responses.

3. Token Generation

Generates a GitHub App token for authenticated API operations.

4. Instruction Extraction

Extracts and processes the user's instruction:

  • Removes the @mention from the content
  • Validates input length (max 4000 characters)
  • Sanitizes the input for safe processing

5. AI Response

Sends the processed request to Claude Code with:

  • Repository context
  • User instruction
  • Formatting guidelines for GitHub comments

Implementation Details

Event Triggers

on: issue_comment: types: [created] pull_request_review_comment: types: [created] pull_request_review: types: [submitted] issues: types: [opened]

Conditional Execution

if: | ( (github.event_name == 'issue_comment' && contains(github.event.comment.body, '@efp-dev-ops')) || (github.event_name == 'pull_request_review_comment' && contains(github.event.comment.body, '@efp-dev-ops')) || (github.event_name == 'pull_request_review' && contains(github.event.review.body, '@efp-dev-ops')) || (github.event_name == 'issues' && contains(github.event.issue.body, '@efp-dev-ops')) ) && contains(fromJSON(vars.ALLOWED_USER_LIST), github.actor)

Security Considerations

  • User Authorization: Only authorized users can trigger the workflow
  • Input Validation: User input is validated and truncated to prevent abuse
  • Token Scope: Uses minimal required permissions
  • Error Handling: Continues execution to prevent denial of service

Customization

Changing the Mention Trigger

Update the conditional logic and extraction steps to use a different mention pattern:

# Replace @efp-dev-ops with your desired mention sed 's/@efp-dev-ops/@your-bot-name/g' .github/workflows/ai-on-demand.yaml

Adjusting Input Limits

Modify the input validation step to change the character limit:

# Change the 4000 character limit INSTRUCTION=$(echo "$INSTRUCTION" | head -c YOUR_LIMIT)

Custom Prompts

Modify the direct_prompt section to customize how the AI responds:

direct_prompt: | Your custom system prompt here. **User Request:** ${{ steps.extract-instruction.outputs.instruction }} Your custom instructions for response format.

Troubleshooting

Common Issues

  1. Workflow not triggering

    • Verify the user is in ALLOWED_USER_LIST
    • Check that the mention format is correct
    • Ensure the workflow file is in .github/workflows/

  2. Authentication errors

    • Verify GitHub App credentials
    • Check Claude Code OAuth token
    • Ensure proper permissions are set

  3. Response errors

    • Check timeout settings
    • Verify network connectivity
    • Review Claude Code service status

Debug Steps

  1. Check the Actions tab for workflow execution logs
  2. Verify environment variables and secrets
  3. Test with a simple request first
  4. Review the extracted instruction output

This workflow provides a powerful way to integrate AI assistance directly into your GitHub development process, making it easy for team members to get contextual help when needed.