SwiftGuard

AI-powered PR review agent for Swift/SwiftUI projects. Runs as a GitHub Action on every pull request.

What it does

SwiftGuard acts as an automated senior engineer, analyzing your PRs across 7 phases:

Phase	What	LLM needed?
1. Context Gathering	Git diff, dependencies, existing tests, commits	No
2. Static Analysis	PR size, commit lint, naming, docs, UI elements	No
3. LLM Analysis	Architecture, security, performance, breaking changes	Yes
4. Test Generation	XCUITests with compiler feedback loop	Yes
5. Test Execution	Run tests on simulator, capture screenshots	No
6. Visual Analysis	Vision LLM reviews screenshots	Yes
7. Report	PR comment, reviewer assignment, labels	No

Quick start

Use in your repo (GitHub Action)

1. Add ANTHROPIC_API_KEY (or OPENROUTER_API_KEY) to your repo secrets

2. Create .github/workflows/swiftguard.yml:

name: SwiftGuard
on:
  pull_request:
    types: [opened, synchronize, reopened]

permissions:
  pull-requests: write
  contents: read

jobs:
  review:
    runs-on: macos-15
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - uses: issyugk3/Hackathon-XCUItestAgent@main
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}

3. Open a PR — SwiftGuard posts a review comment automatically.

Run locally

# Build
swift build

# Dry run (no API key needed)
.build/debug/swiftguard /path/to/project --mode local --base-sha HEAD~1 --dry-run

# Full run with LLM
export OPENROUTER_API_KEY="sk-or-..."
.build/debug/swiftguard /path/to/project --mode local --base-sha HEAD~1

Configuration

Add .swiftguard.yml to your project root:

llm:
  text:
    provider: openrouter        # anthropic | openai | openrouter | openai-compatible
    model: openai/gpt-4o
  vision:
    provider: openrouter
    model: openai/gpt-4o

testing:
  scheme: MyApp
  simulator: "iPhone 16 Pro"
  enable_execution: false       # set true to run XCUITests

code_quality:
  commit_format: conventional
  architecture: mvvm
  naming:
    enforce_swift_guidelines: true
  max_pr_size:
    small: 100
    medium: 300
    large: 500

team:
  auto_assign_reviewers: true
  module_owners:
    - path: "Sources/Auth/"
      owners: ["@alice"]
  require_doc_update: true
  auto_labels: true

What SwiftGuard catches

• Security: hardcoded secrets, force unwraps, insecure network calls
• Performance: retain cycles, unnecessary re-renders, main thread blocking
• Code quality: naming violations, commit message format, missing docs
• Architecture: pattern violations, breaking changes, impact analysis
• Visual: layout issues, accessibility, consistency (via screenshots)
• Team: auto-assign reviewers, PR size labels, split suggestions

Supported LLM providers

Provider	Text	Vision	Env var
Anthropic	Yes	Yes	ANTHROPIC_API_KEY
OpenAI	Yes	Yes	OPENAI_API_KEY
OpenRouter	Yes	Yes	OPENROUTER_API_KEY
OpenAI-compatible	Yes	No	LLM_API_KEY

CLI options

swiftguard [project-path] [options]

Options:
  --mode <mode>           github-action or local (default: local)
  --base-sha <sha>        Base commit for diff
  --pr-number <number>    PR number (github-action mode)
  --repo <owner/repo>     Repository (github-action mode)
  --simulator <id>        Simulator device ID
  --config <path>         Path to .swiftguard.yml
  --dry-run               Skip LLM phases, static analysis only

License

MIT