Spec Drift Detector

AI-assisted development accelerates builds but also accelerates drift. Cursor, Copilot, and similar tools optimise for working code — not for your product intent. This skill compares what was specced against what was built, finds the gaps, and tells you what to do about them.

---

Context

Spec drift is not always a mistake. This skill distinguishes between:

Negative drift: build is missing something from the spec, or contradicts it

Positive drift: build adds something better than what was specced

Scope creep: build includes things that were explicitly out of scope

Clarification drift: spec was ambiguous and engineering made a reasonable assumption

All four types need to be documented. Only the first three need action.

---

Step 1 — Gather inputs

The original spec and the current state of the build (PR description, demo notes, QA results, or written description).

Step 2 — Extract requirements from the spec

Number each requirement: R-01, R-02, etc. Preserve GIVEN/WHEN/THEN format if used.

Step 3 — Extract what the build actually does

Number each behaviour: B-01, B-02, etc.

Step 4 — Run the comparison

For every spec requirement, find its match in the build: ✅ Fully implemented, ⚠️ Partially built, ❌ Not implemented, 🔄 Implemented differently. Then check build for behaviours with no matching spec requirement. Assign severity: Critical / High / Medium / Low.

Step 5 — Generate the drift report

Summary with counts, overall status (Green/Amber/Red), requirement-by-requirement breakdown, items requiring action (must fix / should fix / document and accept), recommended spec updates, and open questions for engineering.

Quality check before delivering

Suggested next step: Share this report in your next sprint standup. "Must fix" items should become sprint tasks immediately. Update the spec with any accepted improvements before closing the sprint.