Mr. Code Fixer 🤖

An AI-powered GitHub bot that automatically analyzes and fixes issues in your repositories. Download, configure, and let it run autonomously to maintain your projects.

What Does It Do?

Mr. Code Fixer is an autonomous bot that:

• 🔍 Monitors your GitHub repositories for open issues
• 🧠 Uses AI to understand what needs to be fixed
• 💡 Asks clarifying questions when uncertain
• ✏️ Creates fixes and opens pull requests
• 🧪 Runs tests before creating PRs
• 📊 Tracks API costs and session metrics
• ✅ Closes issues when confident about the solution

Think of it as a tireless contributor that works 24/7 to help maintain your codebase.

Features

• Smart Issue Processing: Filters out vague issues, duplicates, and PRs automatically
• Multi-AI Support: Choose between ChatGPT (OpenAI), Grok (xAI), or Ollama (local)
• Confidence-Based Decisions: High confidence fixes auto-close issues; uncertain ones ask questions
• Test Execution: Automatically detects and runs tests (Go, Node.js, Python, Rust, Java, PHP)
• Cost Tracking: Shows estimated costs before processing multiple issues
• Session Analytics: Tracks API calls, costs, PRs created, and questions asked
• Beautiful UI: Colored output with emojis and progress indicators

Quick Start

1. Download

Get the latest release for your platform from the releases page:

• mr-code-fixer_Windows_x86_64.zip - Windows
• mr-code-fixer_Linux_x86_64.tar.gz - Linux (Intel/AMD)
• mr-code-fixer_Linux_arm64.tar.gz - Linux (ARM)
• mr-code-fixer_Darwin_x86_64.tar.gz - macOS (Intel)
• mr-code-fixer_Darwin_arm64.tar.gz - macOS (Apple Silicon)

2. Setup

Run the bot for the first time:

./mr-code-fixer

It will guide you through interactive setup:

• GitHub Repository: Which repo should the bot help fix?
• GitHub Token: Your personal access token (see below)
• AI Service: Choose ChatGPT, Grok, or local Ollama
• Working Directory: Where to clone repos (defaults to ~/.mr-code-fixer/workspace)

Configuration is saved in ~/.mr-code-fixer.json for future runs.

3. Run

The bot will:

1. Show all open issues
2. Let you select which one to fix
3. Analyze the issue with AI
4. Either ask questions (if uncertain) or create a PR (if confident)

Use Cases

Personal Projects

Run manually whenever you want help with issues:

./mr-code-fixer

Automation

Set up the bot to run periodically:

Windows (Task Scheduler):

# Run every 6 hours
$trigger = New-ScheduledTaskTrigger -Once -At (Get-Date) -RepetitionInterval (New-TimeSpan -Hours 6)
$action = New-ScheduledTaskAction -Execute "C:\path\to\mr-code-fixer.exe"
Register-ScheduledTask -TaskName "Mr Code Fixer" -Trigger $trigger -Action $action

Linux/macOS (cron):

# Add to crontab (runs every 6 hours)
0 */6 * * * cd /path/to && ./mr-code-fixer

Docker:

FROM golang:1.21-alpine
WORKDIR /app
COPY . .
RUN go build -o mr-code-fixer
CMD ["./mr-code-fixer"]

Dedicated Bot Account

For a true "bot experience":

1. Create a new GitHub account (e.g., my-code-fixer-bot)
2. Set up a nice bot profile (avatar, bio: "🤖 Automated code fixer")
3. Generate a PAT from the bot account
4. Invite the bot as a collaborator to your repos
5. Run Mr. Code Fixer with the bot's token

Now all PRs and comments will appear from the bot account!

Configuration

GitHub Personal Access Token

Create a token at https://github.com/settings/tokens with these permissions:

Fine-grained token (recommended):

• Repository access: Select repositories you want the bot to help with
• Permissions:
  • Contents: Read and write
  • Issues: Read and write
  • Pull requests: Read and write
  • Metadata: Read-only

Classic token:

• repo (full control)

AI Services

Choose one of three AI providers:

1. ChatGPT (OpenAI)

• Get API Key: https://platform.openai.com/api-keys
• Models: gpt-4o, gpt-4-turbo, gpt-3.5-turbo
• Cost: Pay-per-use (check OpenAI pricing)

2. Grok (xAI)

• Get API Key: https://x.ai
• Models: grok-3, grok-4-fast-reasoning, grok-code-fast-1
• Cost: Pay-per-use (check xAI pricing)

3. Ollama (Local)

• Install: https://ollama.ai
• Models: llama2, codellama, deepseek-coder (free, runs on your machine)
• Cost: Free, but uses your compute resources
• Setup: ollama pull codellama then select in the bot

How The Bot Thinks

Confidence-Based Decisions

The bot assesses its confidence before acting:

High Confidence ✅

• Creates PR with the fix
• Adds comment to issue
• Automatically closes the issue

Medium/Low Confidence ⚠️

• Creates PR with warnings
• Leaves issue open for human review

Needs More Info ❓

• Posts questions as issue comments
• Waits for human clarification
• Does NOT create a PR

Branch Naming

The bot creates descriptive branches:

• fix/1-app-crashes-on-startup
• fix/23-typo-in-documentation
• fix/5-where-is-documentation

Example Workflow

Issue #5: "App crashes on startup"
  ↓
Bot analyzes code + issue description
  ↓
AI analyzes and creates fix
  ↓
Bot detects test command (e.g., "go test")
  ↓
Runs tests after applying changes
  ├─ Tests pass → Creates PR, closes issue
  └─ Tests fail → Rolls back, reports error

Session Analytics

After processing issues, the bot shows a summary:

╔══════════════════════════════════════════════════════════════╗
║                     📊 Session Summary                        ║
╚══════════════════════════════════════════════════════════════╝

⏱  Duration: 2m 34s
🤖 AI Calls: 5
💰 Estimated Cost: $0.005
📝 Issues Handled: 3
🔀 PRs Created: 2
❓ Questions Asked: 1

When fixing multiple issues, you'll see cost estimates first:

💰 Estimated cost for 10 issues: ~$0.015 (15 AI calls @ $0.001 each)
⚠️  Note: Processing multiple issues will incur API costs
Fix all 10 issues? (yes/no) [no]:

Advanced Usage

Configuration File

The bot saves settings in ~/.mr-code-fixer.json. You can edit this directly:

{
  "github_token": "ghp_xxxxx",
  "repo_owner": "yourusername",
  "repo_name": "yourrepo",
  "ai_service": "grok",
  "ai_api_key": "xai-xxxxx",
  "ai_model": "grok-code-fast-1",
  "work_dir": "/home/user/.mr-code-fixer/workspace"
}

Multiple Repositories

To use the bot with multiple repos, either:

• Run it interactively and change the repo each time
• Create separate config files and use: ./mr-code-fixer --config /path/to/config.json
• Set up separate bot instances with different working directories

Supported Test Frameworks

Mr. Code Fixer automatically detects and runs tests for:

Language/Framework	Detection	Command
Node.js	package.json	npm test
Go	go.mod	go test ./...
Python	requirements.txt, setup.py	python -m pytest
Rust	Cargo.toml	cargo test
Java (Maven)	pom.xml	mvn test
Java (Gradle)	build.gradle	gradle test
PHP	composer.json	php vendor/bin/phpunit

If no tests are found, the bot proceeds without test validation and notes this in the PR.

Building From Source

Requirements

• Go 1.21 or higher
• Git

Build

git clone https://github.com/pefman/Mr-Code-Fixer.git
cd Mr-Code-Fixer
go build -o mr-code-fixer

Create Release

git tag -a v0.2.0 -m "Release v0.2.0"
git push origin v0.2.0
goreleaser release --clean

Limitations & Considerations

• Quality depends on AI: The bot is only as good as the AI model you choose
• Complex issues: May require human intervention or clarification
• No testing: The bot cannot run tests (yet) - always review PRs before merging
• API limits: Respects GitHub API rate limits (5000 requests/hour for authenticated)
• Cost awareness: ChatGPT and Grok are paid services - monitor your usage
• File limit: Analyzes up to 30 most relevant files per issue to keep AI context manageable

Writing Issues the Bot Understands

The bot works best with well-written issues. Here are examples:

✅ Perfect Issue Example

Title: Login button not working in src/components/Auth/LoginForm.tsx

Description:
The login button on the login page doesn't submit the form when clicked.

File: src/components/Auth/LoginForm.tsx
Line: Around line 45

Expected: Clicking "Login" should call handleSubmit() and send credentials to API
Actual: Nothing happens when clicking the button

Error in console:
TypeError: Cannot read property 'submit' of undefined

Why this works:

• ✅ Mentions exact file path: src/components/Auth/LoginForm.tsx
• ✅ Describes expected vs actual behavior
• ✅ Includes error message
• ✅ Specifies approximate location (line 45)

✅ Good Issue Example

Title: Database connection failing in api/db.go

The app crashes on startup with "connection refused" error.
I think the issue is in api/db.go where we initialize the database pool.

Why this works:

• ✅ Mentions file: api/db.go
• ✅ Clear error description
• ✅ Hints at the problem area

⚠️ Okay Issue Example

Title: Typo in documentation

README.md has wrong command - it says `npm start` but should be `npm run dev`

Why this works:

• ✅ Mentions file: README.md
• ✅ Clear what needs to change
• ⚠️ Simple fix, bot will handle confidently

❌ Poor Issue Example

Title: App doesn't work

Something is broken. Please fix it.

Why this fails:

• ❌ No file mentioned
• ❌ No error description
• ❌ No context about what "doesn't work" means
• Bot response: Will ask clarifying questions in the issue

Best Practices for Issues

1. Mention files explicitly: Use backticks for file paths: src/utils/helper.js
2. Include error messages: Copy-paste actual errors from console/logs
3. Describe expected behavior: What should happen vs what actually happens
4. Add context: Environment, steps to reproduce, related files
5. Use keywords: Words like "login", "database", "api" help the bot find relevant files

How the Bot Finds Files

The bot uses smart file selection:

• Explicit mentions: Files mentioned in the issue get highest priority
• Keyword matching: Finds files with issue keywords in their path
• Relevance scoring: Ranks files by how likely they are related
• Limit: Analyzes top 30 most relevant files (not entire codebase)

Example: Issue mentions "login problem" → Bot prioritizes:

1. Files explicitly mentioned: auth/login.js
2. Files with "login" in path: components/LoginForm.tsx, services/loginService.js
3. Files with "auth" in path: middleware/auth.js
4. Common entry points: index.js, main.go, app.py

Tips for Best Results

1. Write clear issues: Mention file paths and include error messages
2. Start with simple issues: Test the bot on documentation or simple bugs first
3. Review PRs carefully: Always review before merging, especially for critical code
4. Use labels: Consider only letting the bot handle issues labeled "auto-fix" or "good-first-issue"
5. Monitor costs: If using paid AI services, track your API usage
6. One issue per problem: Don't bundle multiple unrelated problems in one issue

Contributing

Contributions welcome! Areas for improvement:

• Add automated testing before creating PRs
• Implement GitHub App for easier installation
• Add webhook listener for real-time responses
• Support for more AI providers
• Better code context understanding
• Issue priority/labeling system

License

MIT License - See LICENSE file for details.

Credits

Built with Go and AI. Zero external dependencies beyond the standard library.

---

Need help? Open an issue or check existing issues for common problems and solutions.