Saturnday

Govern AI-built software from scan to build to repair — all from the terminal. Saturnday runs 48+ deterministic governance checks on Python projects (with additional checks for TypeScript, DevOps files, and project metadata), builds projects from a brief with ticketed execution and per-commit governance, produces governed documents, inspects release artefacts, and auto-fixes findings with evidence.

Requires: pip install saturnday (Python 3.10+)

Verify installation:
CODEBLOCK0

If saturnday is not installed, run:

pip install saturnday

---

Three Modes

Saturnday has three core modes plus document and release modes. Choose based on the task:

Mode	When to use	Command
Scan	Check a skill or repo for issues	INLINECODE1
Guard

Full governance on a git repo | python scripts/guard.py <path> |
| Run | Build a project from a brief | python scripts/run.py <path> --brief "..." |
| Document | Produce a governed document | saturnday start --document --doc-spec spec.yaml |
| Release | Inspect artefacts before publishing | saturnday release-preflight --repo <path> |

---

Mode 1: Scan

Use when: checking an OpenClaw skill for security risks, hallucinated imports, fake tests, or quality issues before installing or publishing.

CODEBLOCK2

Or directly:
CODEBLOCK3

What it checks

• - 19 Python security checks (SQL injection, auth bypass, CSRF, XSS, hardcoded secrets)
• 19 TypeScript security checks
• Hallucinated import detection (packages that don't exist)
• No-assert and fake test detection
• Dead code (cross-file analysis)
• Dependency declaration verification
• Project hygiene (README, LICENSE)
• Dockerfile checks (unpinned images, root user, secrets in build, missing dockerignore)
• GitHub Actions (SHA pinning, broad permissions, pullrequesttarget, plaintext secrets)
• GitLab CI (secrets in YAML, docker-in-docker patterns)
• Jenkins (hardcoded credentials)
• Terraform (hardcoded credentials, public CIDR)
• Kubernetes (privileged containers, unpinned images, secrets in manifests)
• Optional Trivy integration for deep IaC analysis

Output

Disposition: PASS or FAIL
Findings: list with check name, severity, file, line, description
Evidence: timestamped directory with full results

---

Mode 2: Guard

Use when: running full governance on any git repository — before merging, before deploying, or for audit.

CODEBLOCK5

For staged changes only (pre-commit):
CODEBLOCK6

Or directly:
CODEBLOCK7

What it checks

48+ deterministic checks for Python projects: security (SQL injection, XSS, CSRF, auth bypass, hardcoded secrets, WebSocket security, OAuth, token handling, rate limiting, IDOR, user enumeration, cookie security), AI-specific (hallucinated imports, fake tests, dead code, placeholders), quality (syntax, dependencies, Python version compat, code quality, blast radius, project hygiene, typosquat detection). Additional checks activate for DevOps files (Dockerfile, GitHub Actions, GitLab CI, Jenkins, Terraform, Kubernetes, config security) and TypeScript/JavaScript code.

Ratchet baselines

Prevent regressions: CODEBLOCK8

Policy exemptions

Create .saturnday-policy.yaml:

expected_findings:
  - declared_not_installed
  - package_not_importable

---

Mode 3: Run

Use when: building a project from a description. This is the full governed execution pipeline.

CODEBLOCK10

Or interactively:
CODEBLOCK11

Backends

Backend	Value	Auth
Claude Code CLI	INLINECODE7	Claude Pro subscription
Codex CLI

codex-cli | OpenAI subscription | | OpenClaude | openclaude | OpenClaude CLI | | Cursor CLI | cursor-cli | Cursor | | OpenAI API | openai | OPENAIAPIKEY | | Anthropic API | anthropic | ANTHROPICAPIKEY |

What happens during a run

1. 1. Security triage — LLM-powered data flow analysis filters false positives from security findings so you only review what matters
2. Brief enrichment — vague briefs are auto-enriched with technical detail before planning
3. Planning — 3-stage planner generates tickets with acceptance criteria and scope constraints
4. Execution — each ticket executed sequentially, governance checks after every commit
5. Auto-install — dependencies are automatically installed between tickets when pyproject.toml or package.json changes
6. Retry — if governance fails, revert, feed findings back, retry (up to 3 attempts)
7. Ungoverned commit — if a ticket still fails after retries and auto-repair, the code is committed with a [GOVERNANCE: review required] tag so the project stays complete
8. Auto-repair — after all tickets, ungoverned and failed tickets go through the repair pipeline again
9. Definition of done — evaluates whether plan goals are met
10. Review report — for ungoverned tickets: review-required.md with findings, remediation tips, and copy-paste fix prompts for each ticket
11. Progress log — live saturnday-progress.log with ticket outcomes, governance results, and counterfactual messages showing what ungoverned AI coders would have done wrong
12. Evidence — timestamped directory with per-ticket results, governance evidence, analytics

Resume interrupted runs

Saturnday auto-detects prior runs. If you restart saturnday run on a project that already has a ledger, tickets that already passed are skipped automatically. You can also resume explicitly:

CODEBLOCK12

---

Document Mode

Use when: producing governed documents (specs, reports, audits) with section-level checks and cross-section consistency validation.

CODEBLOCK13

The pipeline: parse spec, plan sections, generate each section, run section-level checks (structure, placeholders, citations, evidence coverage), retry failures, run cross-section consistency checks, and produce a signed-off evidence pack.

---

Release Preflight

Use when: inspecting built artefacts (wheels, sdists, npm tarballs, OCI images) before publishing.

CODEBLOCK14

Runs 5 release-security checks: source map blocking, secrets in artefact, internal file blocking, allowlist manifest validation, and release diff comparison. Separate from code governance.

---

When to Use Each Mode

• - "Scan this skill before I install it" → Scan
• "Check this repo for security issues" → Guard
• "Build me a project from this description" → Run
• "Audit this codebase" → Guard
• "Produce a governed document" → Document mode
• "Check this wheel before publishing" → Release preflight
• "Fix the findings" → use saturnday repair --repo <path> directly

Prerequisites

• - pip install saturnday (Python 3.10+)
• Node.js 18+ (for TypeScript checks)
• Git (for Guard and Run modes)
• At least one AI coder backend for Run mode