saturnday 0.3.0

Execution governor for AI-generated code — Guard, Run, and Repair in one package

Saturnday

AI-specific governance for code generated by Claude, Cursor, Copilot, and other AI coding tools.

Your team already uses AI coding tools. Saturnday makes sure what they produce is safe to ship.

What it does

Saturnday runs 18+ automated checks on every PR, catching the things AI gets wrong that humans miss:

• Hallucinated imports — packages that don't exist on npm or PyPI
• Fake tests — assert True, empty test bodies, tests that test nothing
• Leaked secrets — API keys, passwords, tokens hardcoded in generated code
• Syntax errors — code that won't parse (Python and TypeScript/JavaScript)
• Placeholder code — TODO stubs, pass-only functions, NotImplementedError bodies
• Typosquat dependencies — suspicious package names that look like popular libraries
• Prompt injection — system prompt leakage in generated code
• Security vulnerabilities — via Bandit static analysis
• Shell script issues — via ShellCheck integration
• API version mismatches — calling methods that don't exist on the installed library version

Install

pip install saturnday

Usage

Check a PR diff

saturnday check --repo . --diff HEAD~1..HEAD

Scan an OpenClaw skill

saturnday scan --skill ./skills/my-skill/ --output ./results

Scan an entire skills corpus

saturnday scan --openclaw ./openclaw-skills/ --output ./results --format json

Run — governed execution

Turn a brief into governed tickets and execute them with your AI coder.

saturnday plan --brief "Build a calculator skill" --repo . --backend codex-cli
saturnday run --plan plan.json --repo . --backend codex-cli
saturnday resume --plan plan.json --repo . --backend codex-cli

Repair — fix findings automatically

Scan, generate repair tickets, and execute fixes.

saturnday repair --skill . --dry-run
saturnday repair --skill . --backend codex-cli

GitHub Action

Add this to .github/workflows/saturnday.yml in your repo:

name: Saturnday Governance
on:
  pull_request:
    branches: [main]

permissions:
  contents: read
  pull-requests: write

jobs:
  governance:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - uses: actions/setup-python@v5
        with:
          python-version: '3.12'
      - uses: honouralexwill/saturnday/.github/actions/saturnday-check@master

Every PR gets scanned. Findings are posted as a PR comment. The check fails if the disposition is FAIL.

Policy configuration

Create a saturnday-policy.yaml in your repo root to customise check severity:

schema_version: "1.0.0"

checks:
  secrets: error
  hallucinated_imports: error
  fake_tests: warning
  placeholders: info
  syntax: error

scope:
  denied_paths:
    - "vendor/**"
    - "node_modules/**"

How it works

Saturnday treats AI-generated code as untrusted input. Every diff goes through deterministic verification — no LLM calls in the governance pipeline. The checks are fast, reproducible, and auditable.

Results are written to structured evidence packs that can be stored for compliance and audit purposes.

License

MIT