A production-grade AI development governance skill for Claude Code.
Enforces modular architecture, adversarial feedback, patch-first editing, blast radius checks, zero-trust secrets, and context drift prevention via pre-flight anchor checks, session scoping, and re-injection rules.
Personal (all projects):
cp -r code-warden ~/.claude/skills/Per-project:
cp -r code-warden .claude/skills//code-warden
Or just start a coding session — auto-triggers on: "start a new module", "begin coding", "governance check", "load protocol", and more.
Every AI coding session gets:
- A Session Start Hard Gate (outputs architecture state, session scope, and reference file status as its first response — blocks all further output until the user confirms or provides missing info)
- Pre-flight manifests before large code blocks (file, line count, concern, secrets — human-verifiable)
- Blast Radius Checks before any rewrite (what breaks, how to test, how to rollback)
- Human Checkpoints (
[AWAITING CONFIRMATION]) before multi-file or high-volume changes - Drift detection that halts output when the AI guesses syntax, skips safety, or goes monolithic
| File | Covers |
|---|---|
| SKILL.md | Session checklist, quick rules, drift signals |
| CONFIGURE.md | All tunable thresholds with defaults, rationale, and team profiles |
| DECISIONS.md | Decision log — seeded with real entries |
| references/architecture.md | Blueprint Rule, State Update, Re-injection |
| references/safety.md | Blast Radius, Patch-First, Zero-Trust, Dependency Freeze |
| references/cognition.md | Think Before Coding, Don't Guess Syntax, Human Checkpoint |
| references/cleanup.md | Tech Debt format, Test Contract, Decision Log |
| references/anti-drift.md | Pre-Flight Anchor Check, Session Scoping, Drift Trigger |
| examples/governed-session.md | Annotated example session showing all rules firing |
Thresholds (file size limits, checkpoint triggers, pre-flight thresholds) are opinionated defaults tuned for solo developers. See CONFIGURE.md for a full table of what to change and team-size profiles (solo, small team, monorepo, high-security).
v2.2.3 — See SKILL.md metadata for changelog.