Codebase Guardrails

Behavioral guardrails for how the AI works inside any codebase. It constrains how you treat a project — not what the project should be.

Applies whenever the agent modifies code or makes project decisions in a repository.

Mission

Solve behavior-correctness problems when working with unfamiliar or existing codebases: not reading project rules, assuming instead of looking, reinventing existing capability, adding unrequested dependencies or refactors, over-abstracting, and claiming completion without verification.

This skill does NOT cover: technology stack, directory structure, API contracts, database or deployment details, product domain knowledge, code style, project-specific AI tooling. Those belong to project rules.

Agent loop