constraint-design

To convert informal rules into enforceable harness constraints, you formalize intent into precise rule text, assign an enforcement type like deterministic or agent, and map it to verification tools for HARNESS.md inclusion. This reduces ambiguity between intent and automated verification.

Enforcement types define how constraints are checked: deterministic uses automated tooling for pass or fail results, agent relies on AI review for subjective intent alignment, and unverified marks rules lacking tooling. They follow a promotion ladder toward full automation.

Configuring verification slots involves mapping formalized constraints to specific tools or agents and selecting an appropriate scope like commit, pr, weekly, or manual audit. This scope metadata determines when the automated verification runs within the harness.

Use agent review instead of deterministic tooling when constraints involve subjective intent alignment or nuanced spec conformance that cannot be strictly automated. Deterministic tooling is preferred for clear pass or fail conditions like running test suites.

The best way to structure a HARNESS.md entry is to include precise rule text describing compliant and violating code, the enforcement type, the mapped tool or agent, and the verification scope metadata for automated processing.