B-M9QZ-standards-principles-drift-audit-skill
A skill that loops over every normative statement in the project — each standard under docs/planning/standards/ (S####) and each engineering principle in PRINCIPLES.md — and audits the rest of the project (code + docs) for contradictions or drift against it.
Per rule: load the statement, scope it to applies_to.paths where the standard declares one, scan the relevant surface, and report concrete violations (file:line plus why it contradicts or drifts). Likely an LLM-driven semantic audit rather than a deterministic checker — it complements the existing deterministic checks (/project-check, /sdlc:entities-audit), which catch schema/entity drift but not adherence to prose rules.
Motivating example: the S0005 cleanup cited check_entities.py when the real tool is check_entities.ts — exactly the standard-vs-reality drift this skill would surface.
Open questions: per-rule sub-agent fan-out vs a sequential loop; how to bound the scan so it does not re-read the whole repo per rule; output shape (single report vs promoted backlog items or tasks per violation); whether this is /loop over a single-rule audit skill rather than a bespoke looping skill.