GovKit puts your AI coding agent inside a governed system where your architecture contracts, acceptance criteria, and evaluation thresholds are the source of truth — not the model's training data.
# install, apply, calibrate
pip install govkit
govkit apply --agent claude-code --target .
govkit calibrateWorks with Claude Code, GitHub Copilot, and OpenAI Codex · Any language — Python, C#, Java, Go, TypeScript · Python 3.11+ dev-tool requirement only, never a project dependency.
AI coding agents are powerful — but without constraints, they drift. They invent architecture, skip tests, ignore NFRs, and make decisions that belong to your team. GovKit replaces "trust the model" with a governed workflow: your contracts, specs, and CI gates decide what good looks like — every feature, every time.
↑ ASSET B — "Drift vs. Governed" icon trio goes here (drift arrow · spec page · CI shield).
Path-scoped rules load automatically as the agent edits each layer — api, services, ports, adapters, security. It reads your architecture contracts on every turn, not its own assumptions.
Every feature lives under features/<name>/ with five required artifacts: Gherkin acceptance criteria, fully-populated NFRs, schema-validated eval criteria, a plan, and an architecture preflight.
Ships agent skills — /architecture-preflight, /spec-planning, /implementation-plan, /adr-author — that validate features against your contracts and predict FIRST + 7-Virtue scores before any code is written.
govkit doctor checks whether installed governance still fits your repo; govkit validate checks whether features meet the contract. Both run in CI: schema validation, NFR coverage, boundary enforcement, prediction thresholds.
One toolkit for Claude Code, Copilot, and Codex, across FastAPI, .NET, Spring Boot, Fastify, Gin, dbt, and Databricks overlays. Monorepo-aware — one govkit apply per app subdirectory.
For LLM features, add governed model routing (LiteLLM), observability (OpenLLMetry + Langfuse), and evaluation/safety gates (DeepEval, Promptfoo, RAGAS, NeMo Guardrails, Guardrails AI).
↑ ASSET C — six capability glyphs (shown as placeholders) to be delivered as one icon family.
You need Python 3.11+ on your machine (GovKit is a dev tool, never a project dependency) and an AI coding agent — Claude Code, GitHub Copilot, or OpenAI Codex. GovKit ships the governance configuration; you bring the agent.
# 1 — install
pip install govkit
# 2 — from your project root
govkit apply --agent claude-code --target .
# 3 — make the generic defaults match your repo
govkit calibrate
# 4 — commit your governance baseline
git add .govkit CLAUDE.md docs/ governance/ ci/
git commit -m "chore: add govkit governance baseline"govkit calibrate walks you through a 9-step review; prefer to review with the team first? Run govkit calibrate --non-interactive to write a checklist file instead.Upgrading later? pip install --upgrade govkit && govkit upgrade --target . refreshes the files GovKit owns without touching the contracts you've customized.
govkit init my_feature --target .Creates features/my_feature/ from the appropriate starter (Level 4+). At L3 this errors with a pointer to --level 4 — Foundations ships rules and contracts only.
/architecture-preflight my_feature
/adr-author my_feature
/spec-planning my_feature
/implementation-plan my_featureThe agent produces architecture_preflight.md, plan.md, and eval_criteria.yaml — and will not proceed if predicted FIRST or 7-Virtue averages fall below 4.0.
| Step | Claude Code | Copilot | Codex |
|---|---|---|---|
| Architecture preflight | /architecture-preflight my_feature | /architecture-preflight | $architecture-preflight my_feature |
| Spec planning | /spec-planning my_feature | /spec-planning | $spec-planning my_feature |
Copilot infers the feature from context; Codex invokes skills with a $ prefix.
govkit validate --target .Level-aware per-feature compliance — artifact existence, Gherkin structure, NFR coverage, eval-criteria schema, prediction thresholds. No-op at L3.
Backend APIs, CLIs, React or Angular UIs, and governed dbt / Databricks data projects — across Claude Code, GitHub Copilot, and OpenAI Codex. Any project language; GovKit copies Markdown specs, YAML configs, and Gherkin files into your repo.
↑ ASSET G — badge strip (live shields.io or brand chips). TODO: confirm default branch for raw links.
Install GovKit, apply it to your repo, and let your contracts — not the model — decide what good looks like.
pip install govkitFree and open source. Want help rolling governed AI delivery out across your teams? Talk to Accelerated Innovation.