AI Coding Guardrails Kit
Policy and instruction templates that keep AI coding assistants on track.
A starter kit for teams adopting AI coding assistants: versioned instruction files, policy templates, a linter that checks them for contradictions and a CI step that flags risky AI-generated changes for human review.
- Role
- Author
- Started
- License
- Apache-2.0
About the project
Teams rarely fail with AI coding tools because the models are weak. They fail because nobody wrote down what the assistant is allowed to do. The Guardrails Kit turns that tacit knowledge into versioned, reviewable files.
What is inside
AGENTS.mdand Copilot instruction templates for common stacks.- Policy packs for secrets, licences, dependencies and data protection.
- A CLI that lints instruction files for conflicts, vague rules and stale paths.
- A GitHub Action that labels pull requests touching sensitive areas.
Status
The kit is experimental. Templates are battle-tested in client trainings, while the CLI API may still change.
Features
Policy packs
Ready-made rules for secrets, licences, dependencies and personal data.
Instruction templates
AGENTS.md and Copilot instructions for TypeScript, Java and Python projects.
Instruction linter
Finds contradictions, vague wording and references to files that no longer exist.
CI gate
Flags AI-assisted pull requests that touch sensitive code for mandatory human review.