- Context
-
Requirement statements leak implementation detail. Candidates: strengthen the propose skill, add a validation rule, or add a reviewer.
- Options
-
-
Strengthen the propose skill’s Rules with a self-check.
-
Good: catches the leak where requirements are authored; cheap.
-
Bad: advisory — the authoring agent must apply it.
-
-
A validation or lint rule.
-
Good: automatic and always-on.
-
Bad: "what not how" is semantic, so no structural linter can catch it.
-
-
A dedicated review skill.
-
Good: an independent check.
-
Bad: another skill and step for a one-bullet problem.
-
-
- Decision
-
Strengthen the propose skill’s Rules with a self-check that references the requirement guidance’s "Solution prescription" list.
- Rationale
-
Semantic leakage cannot be linted; the cheapest effective lever is a self-check at authoring time, reusing the flag list the guidance already defines.
- Consequences
-
Reduces but does not eliminate leakage; the human accept gate and the archive cold reviewer remain the backstops.
DEC-012 — Enforce what-not-how at propose, not in validation