AI Workflow Automation

🤖 AI-powered GitHub workflow automation featuring advanced PR review with GitHub's native resolvable suggestions. Intelligent, confidence-based feedback with one-click code fixes and comprehensive workflow capabilities.

✨ Key Features

🔍 AI PR Review (Current Focus)

• 🎯 GitHub Native Resolvable Suggestions: High-confidence AI recommendations (≥95%) become one-click applicable code changes
• 📊 Confidence-Based Intelligence: Multi-factor scoring algorithm evaluates suggestions based on issue severity, static analysis, code context, and historical patterns
• ⚡ Smart Rate Limiting: Maximum 5 resolvable suggestions per PR to prevent cognitive overload
• 🎚️ Graduated Response System: Different presentation formats based on confidence levels (95%+ resolvable, 80-94% enhanced, 65-79% informational)
• 🛡️ Comprehensive Analysis: Security, performance, quality, and architectural review capabilities

🚀 Workflow Automation Platform

• 🔧 Configurable Integration: Customize AI models, parameters, and workflow triggers
• 📈 Extensible Architecture: Built to support multiple AI-powered workflow automations
• 🎯 Information Dense Keywords: Standardized AI command vocabulary for consistent task execution
• ⚙️ Multi-Model Support: Works with OpenRouter's extensive AI model ecosystem

🚀 Quick Start

Step 1: Add Secrets

Add your OpenRouter API key to your repository secrets:

• Go to your repository → Settings → Secrets and variables → Actions
• Add OPENROUTER_API_KEY with your OpenRouter API key

Step 2: Create Workflow

Create .github/workflows/ai-review.yml in your repository:

name: AI PR Review

on:
  pull_request:
    types: [opened, labeled]

permissions:
  contents: read
  pull-requests: write

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - name: Checkout code
        uses: actions/checkout@v4

      - name: AI PR Review with Resolvable Comments
        uses: stillrivercode/[email protected]
        with:
          github_token: ${{ secrets.GITHUB_TOKEN }}
          openrouter_api_key: ${{ secrets.OPENROUTER_API_KEY }}
          model: 'google/gemini-2.5-pro'  # or any OpenRouter model
          review_type: 'full'
          post_comment: true  # Enable resolvable comments

Step 3: Set Up Labels (Optional)

Install standardized GitHub labels for AI workflows:

# Quick setup - download and run the label setup script
curl -s https://raw.githubusercontent.com/stillrivercode/stillriver-ai-workflows/main/scripts/setup-labels.sh | bash

Or add a workflow file .github/workflows/setup-labels.yml:

name: Setup Labels
on:
  workflow_dispatch: # Manual trigger

jobs:
  setup-labels:
    runs-on: ubuntu-latest
    steps:
      - name: Setup AI Workflow Labels
        run: |
          curl -s https://raw.githubusercontent.com/stillrivercode/stillriver-ai-workflows/main/scripts/setup-labels.sh | bash
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

This creates labels like ai-task, ai-completed, ai-reviewed, and others that enhance the AI workflow experience.

Step 4: Open a Pull Request

The action will automatically:

• ✅ Analyze your code changes with AI
• ✅ Post intelligent, confidence-based feedback
• ✅ Create resolvable suggestions for high-confidence recommendations
• ✅ Provide one-click code fixes directly in GitHub

AI Resolvable Comments

This action features an advanced AI review system that generates GitHub's native resolvable suggestions, transforming AI feedback into actionable, one-click applicable code changes.

Key Features

• 🎯 Confidence-Based Suggestions: Multi-factor scoring algorithm evaluates Issue Severity (40%), Static Analysis (30%), Code Context (20%), and Historical Patterns (10%)
• ✨ Native GitHub Integration: High-confidence suggestions (≥95%) become GitHub's resolvable suggestions with one-click application
• 📊 Intelligent Rate Limiting: Maximum 5 resolvable suggestions per PR to prevent cognitive overload
• 🎚️ Graduated Response: Different presentation formats based on confidence levels
• ⚙️ Configurable Inline Comments: Enable/disable GitHub's native resolvable suggestions via environment variables

Confidence Thresholds

Confidence	Type	Description
≥95%	🔒 Resolvable	Critical issues that become GitHub's native resolvable suggestions
80-94%	⚡ Enhanced	High-confidence recommendations with detailed context
65-79%	💡 Regular	Medium-confidence informational comments
<65%	ℹ️ Suppressed	Low-confidence suggestions aggregated into summary

Configuration

The AI resolvable comments system can be configured via environment variables and repository settings:

Environment Variables

• AI_ENABLE_INLINE_COMMENTS - Enable/disable GitHub's native resolvable suggestions (default: true)
  • true: High-confidence suggestions become resolvable with one-click application
  • false: All suggestions use enhanced format without inline resolution
• AI_REVIEW_RATE_LIMIT_MINUTES - Rate limit between AI reviews (default: 1 minute)
• AI_MODEL - AI model to use for analysis (default: google/gemini-2.5-pro)

Repository Variables

Set these in your repository's Settings → Secrets and variables → Actions → Variables:

• AI_ENABLE_INLINE_COMMENTS - Repository-level control over inline comments
• AI_REVIEW_RATE_LIMIT_MINUTES - Repository-level rate limiting configuration
• AI_MODEL - Repository-level AI model selection

Usage Examples

# Default behavior (inline comments enabled)
npm run ai-review-resolvable

# Disable inline comments
AI_ENABLE_INLINE_COMMENTS=false npm run ai-review-resolvable

# Use different model with inline comments disabled
AI_ENABLE_INLINE_COMMENTS=false AI_MODEL=anthropic/claude-3.5-sonnet npm run ai-review-analyze

# Direct script access
./scripts/ai-review-resolvable.sh analyze 123
AI_ENABLE_INLINE_COMMENTS=false ./scripts/ai-review-resolvable.sh analyze 123

Available Commands

# Complete AI review workflow with resolvable comments
npm run ai-review-resolvable

# Analyze code changes and generate suggestions
npm run ai-review-analyze

# Demonstration of suggestion formatting
npm run ai-review-demo

# Direct script access for PR analysis
./scripts/ai-review-resolvable.sh analyze --pr-number=123

Inputs

See action.yml for a full list of inputs. For details on how to use custom_review_rules and understanding the prompt structure, see docs/prompt-template-structure.md.

Required Parameters

• github_token: Your GitHub token for API access
• openrouter_api_key: Your OpenRouter API key for AI model access

Comment Posting

By default, the action automatically posts review comments to your PR.

Set post_comment: false if you only want to use the review content via outputs without posting a comment. This allows for maximum flexibility in how you handle the review content.

github_token Permissions

This action requires the pull-requests: write permission to post review comments. You can grant this permission in your workflow file:

permissions:
  pull-requests: write

How It Works

Input Types Note

While the input descriptions in action.yml use semantic types like number or string for clarity, all GitHub Actions inputs are received as strings. The action handles the necessary parsing and validation internally. For example, max_tokens and temperature are parsed from strings to numbers with appropriate validation.

AI Review Types

The action supports different review types that tailor the AI's focus:

• full (default): Comprehensive review covering bugs, improvements, security, and code style
• security: Focused on identifying security vulnerabilities
• performance: Focused on identifying performance issues

Review Deduplication

The action automatically prevents duplicate reviews by checking for existing AI reviews on the pull request. If an AI review (containing "## 🤖 AI Review by") already exists from the github-actions[bot] user, the action will skip the review and set review_status to skipped.

AI Review Format

AI reviews with resolvable comments feature:

• Resolvable Suggestions: High-confidence suggestions (≥95%) appear as GitHub's native resolvable suggestions with one-click application (configurable)
• Enhanced Comments: Medium-high confidence suggestions (80-94%) provide detailed context and rationale
• Graduated Response: Different presentation formats based on confidence levels
• Rate Limiting: Maximum 5 resolvable suggestions per PR to prevent cognitive overload
• Summary Reports: Comprehensive analysis with categorized suggestions and confidence metrics
• Flexible Configuration: Inline comments can be disabled, converting resolvable suggestions to enhanced format

Review Status

The action sets the review_status output to indicate the result:

• success: Review was generated and posted successfully
• skipped: Review was skipped (duplicate review exists, no changed files, or no files after filtering)
• failure: An error occurred during the review process

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for more information.