Slay Check

Slay Check — AI-powered code review for GitHub Actions and local development

TL;DR — Local (fastest): pip install this repo → slay-check setup → ollama pull llama3.2 → slay-check quick. GitHub PRs: copy a workflow from examples/github-workflows/, add your AI secret, open a PR → one review comment.

Slay Check gives you a single, comprehensive AI review comment on every pull request. It supports multiple AI providers (OpenAI, Anthropic, Google Gemini, Perplexity, Ollama, GitHub Models, and Custom HTTP) with configurable criteria: problem analysis, algorithm review, complexity, and risk evaluation.

Features

• Multi-AI Support: OpenAI, Anthropic, Google Gemini, Perplexity, Ollama (local/free), GitHub Models (free), Custom HTTP
• Free local reviews: Use Ollama (no API key, runs on your machine) or GitHub Models (free with your GitHub account)
• GitHub Actions Ready: Automated PR reviews
• Single comment: Full review (summary + per-file details) in one PR comment
• CLI Interface: Local development support
• MCP server (optional): pip install "slay-check[mcp]" — use slay-check-mcp with Claude Desktop, Cursor, and other MCP clients (docs/mcp.md)
• Assistant skill packs: marketplace/ — Claude SKILL.md and ChatGPT Custom GPT instructions for publishing
• Configurable Analysis: Customize review criteria
• Comprehensive Scoring: Detailed feedback with actionable suggestions
• Security Focused: Built-in security and risk assessment

Quick Start

GitHub Actions (fastest path)

1. Copy a workflow template into your repo:
   • Copy one file from examples/github-workflows/ → .github/workflows/ in your repository.
2. Add the required secret for the template you chose (Settings → Secrets and variables → Actions).
3. Open a PR — Slay Check runs and posts one comment with the full review.

Slay Check is installed from this public repository, so you do not need to create a GitHub PAT just to use the workflow templates. GitHub Actions provides GITHUB_TOKEN automatically for PR metadata and review comments.

Workflow templates (copy-paste):

Workflow file	Provider	Secret required
slay-check-anthropic.yml	Anthropic (Claude)	ANTHROPIC_API_KEY
slay-check-openai.yml	OpenAI (GPT)	OPENAI_API_KEY
slay-check-google.yml	Google AI (Gemini)	GOOGLE_AI_API_KEY
slay-check-perplexity.yml	Perplexity AI	PERPLEXITY_API_KEY
slay-check-custom-http.yml	Custom HTTP endpoint	SLAY_CHECK_AI_TOKEN (optional)

For more options (run from a fork, environment-specific or label-based workflows), see docs/github-action.md and examples/github-workflows/.

Pip cache
Workflows use cache: 'pip' so dependencies are reused between runs. This shortens job time and does not increase GitHub Actions cost (you pay by runner minutes; caching typically reduces them).

Where the review appears
Slay Check posts one comment per PR. By default it’s an issue comment (main conversation under "Conversation"). To use a PR review comment instead, set use_issue_comments: false in slay-check.yaml.

Example review comment (click to expand)

## Code Review Summary

**Overall Score:** 7.5/10

**Files Reviewed:** 3
**Issues Found:** 2

### Issues by Severity
- **Medium:** 1
- **Low:** 1

### Top Suggestions
- Consider using a constant for the retry limit
- Add input validation for the timeout parameter

---

### Per-file details

#### src/client.py (score: 7.0/10)

The HTTP client handles retries but does not validate timeout values.

**Issues:**
- **[medium]** performance (line 42): Unbounded retry loop may cause hangs
  - Suggestion: Add a max-retry constant and backoff strategy

**Suggestions:**
- Extract magic numbers into named constants

Local — minimal effort

Step	Command
1. Install	pip install git+https://github.com/prathameshap/slay-check.git
2. One-line config (Ollama — no API key)	slay-check setup
3. Pull a model once	ollama pull llama3.2
4. Review your working tree	slay-check quick (same as slay-check review --local)

Run slay-check with no arguments for a reminder. Use slay-check config init --quick for the same default YAML without prompts.

PyPI: not published yet — install from GitHub as above.

Copy .env.example to .env if you prefer env vars over editing YAML.

Alternatives (same idea, more manual): set SLAY_CHECK_AI_PROVIDER yourself — see the table below. For MCP (Cursor / Claude Desktop), see docs/mcp.md.

Method	Notes
Ollama	Free, offline — slay-check setup already sets ai_provider: ollama
GitHub Models	Free with a PAT — slay-check setup --provider github then set SLAY_CHECK_AI_TOKEN

From VS Code or Cursor — run the same commands in the integrated terminal, or add MCP (slay-check-mcp) per docs/mcp.md. Optional: copy .env.example → .env.

Windows (PowerShell) for tokens:

$env:SLAY_CHECK_AI_TOKEN = "your-key"
slay-check quick

Configuration

Create slay-check.yaml in your repository (or use environment variables).

Minimal configuration (recommended)

Use a preset to keep it simple:

# One of: full (default), standard, minimal, security, performance
review_preset: security

Preset	What's reviewed
full	Everything (problem, algorithm, best approaches, complexity, risk, security, performance)
standard	Same as full but skip problem analysis
minimal	Security + performance only
security	Security + risk only
performance	Algorithm, best approaches, complexity, performance

Or pick specific areas with a list (only these are enabled):

review_focus:
  - security
  - performance
  - complexity

Allowed focus names: problem, algorithm, best_approaches, complexity, risk, security, performance.

Environment variables: SLAY_CHECK_REVIEW_PRESET=security or SLAY_CHECK_REVIEW_FOCUS=security,performance (comma-separated).

Advanced: set each criterion explicitly with review_criteria:

# AI Provider
ai_provider: openai  # openai, anthropic, google, perplexity, ollama, github, custom_http
ai_model: gpt-4o     # e.g. gpt-4o, claude-sonnet-4-6, gemini-2.5-flash, llama3.2, sonar-pro

# Analysis Criteria (optional; use preset or focus above for simplicity)
review_criteria:
  analyze_problem: true
  algorithm_analysis: true
  best_approaches: true
  complexity_analysis: true
  risk_assessment: true
  security_review: true
  performance_review: true

# File Filtering
exclude_patterns:
  - "*.md"
  - "*.txt"
  - "vendor/*"
  - "node_modules/*"
  - "__pycache__/*"

# Performance Limits
max_file_size: 10000
max_files_per_pr: 50

# Review Thresholds
min_score_threshold: 6.0
critical_issues_limit: 3

For full configuration options (providers, custom HTTP, etc.), see docs/github-action.md.

AI Providers

Provider	Models / endpoints (examples)	Best For
OpenAI	GPT-4 family (e.g. gpt-4o), or any compatible chat/completions model	General code review
Anthropic	Claude family (e.g. claude-sonnet-4-6)	Security and risk-focused analysis
Google AI	Gemini family (e.g. gemini-2.5-flash)	Performance / efficiency analysis
Perplexity	sonar, sonar-pro, sonar-deep-research, sonar-reasoning-pro	Fast, grounded code review
Ollama	Any local model (llama3.2, codellama, mistral, deepseek-coder, …). No API key, no cost.	Free local reviews, air-gapped environments
GitHub Models	gpt-4o, Llama-3.1-8B-Instruct, Mistral-large, … Free with a GitHub account. Uses your GitHub PAT.	Free cloud reviews for GitHub users
Custom HTTP	Any HTTP endpoint (including Cursor-backed gateways). Set ai_provider: custom_http (aliases: http, cursor)	Enterprise / internal models and gateways

Documentation

Where to go next:

Goal	Doc
Run in GitHub Actions	GitHub Actions
Run locally (CLI)	Local Development
Full setup (secrets, keys, troubleshooting)	Setup Guide
Customize review (presets, providers)	Configuration below
Use project icon and social copy	Branding and Assets
Fork or contribute	CONTRIBUTING.md

More: AI Response Control · Forking Guide · Branding and Assets

Development

# Clone and install
git clone https://github.com/prathameshap/slay-check.git
cd slay-check
pip install -e ".[dev]"

# Run tests (all providers mocked; no API keys needed)
pytest

# Build package
python -m build

Testing all services

1. Unit tests (no real APIs)
Uses mocks for every provider. No tokens required.

pip install -e ".[dev]"
export SLAY_CHECK_AI_TOKEN=dummy
export SLAY_CHECK_GITHUB_TOKEN=dummy
pytest tests/ -v

• All tests: pytest tests/ -v
• AI providers only: pytest tests/test_ai.py -v
• Config only: pytest tests/test_config.py -v
• CLI only: pytest tests/test_cli.py -v
• With coverage: pytest tests/ --cov=slay_check --cov-report=term-missing

2. Lint (same as CI)

black --check slay_check tests
isort --check-only slay_check tests
flake8 slay_check tests

3. Manual test against real APIs
To hit real OpenAI/Anthropic/Google/Perplexity/HTTP endpoints, run the CLI with real tokens and --dry-run so nothing is posted to GitHub:

export SLAY_CHECK_AI_TOKEN="your-openai-key"
export SLAY_CHECK_GITHUB_TOKEN="your-github-pat"
# Optional: SLAY_CHECK_AI_PROVIDER=anthropic (default: openai)

slay-check review --repo owner/repo --pr <PR_NUMBER> --dry-run --verbose

Repeat with different SLAY_CHECK_AI_PROVIDER (and corresponding token) to test each provider. For custom HTTP, set SLAY_CHECK_AI_PROVIDER=custom_http and SLAY_CHECK_AI_BASE_URL to your endpoint; the token is sent as Authorization: Bearer <token>.

Performance

• Startup Time: <1 second
• Small PR (1-5 files): <90 seconds
• Medium PR (5-15 files): <4 minutes
• Large PR (15+ files): <12 minutes
• Memory Usage: <500MB peak

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines, including our branching strategy (single main branch, all changes via PRs). By participating, you agree to our Code of Conduct.

Quick Contribution Steps

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests
5. Submit a pull request

License

Slay Check is licensed under AGPL-3.0 for v1.0.0 and earlier. Redistribution must be free of charge. Versions after 1.0.0 may use different terms.

Security

Please report security issues as described in SECURITY.md.

Support

• Something wrong? See Troubleshooting in the setup guide.
• Issues: GitHub Issues
• Discussions: GitHub Discussions

---

Slay Check — built for the developer community