Stet

Stet is change control for AI coding behavior. It tells you whether a model, config, skill, or workflow change is safe to ship by replaying real repo work and scoring the output on correctness and quality.

Use this skill as the single entry point and treat it as the agent-facing optimization interface. Use the object model and canonical read path, then pick the cheapest flow that preserves the decision semantics the operator needs. Default to the manifest-backed rules flow when the change is a shared instruction, policy, harness, docs, or model rollout rather than a throwaway directional check.

Agent-Facing Optimization Interface

Stet is the Evaluation Function plus the context that teaches the agent how to use it effectively. The coding agent is the optimizer. The skill's job is to keep that optimizer inside a bounded search loop over real repo work.