Guard against scope creep
Creep is rarely one big overreach; it is ten small "while I'm here" additions. This skill turns every addition into a visible, priced item rather than an absorbed cost of the deliverable.
Procedure
Keep the frozen requirement list open beside the work so you can test each impulse against it.
Classify everything you are tempted to add:
in-scope— directly required by an existing criterion; just do it.adjacent— needed to make a criterion pass but not itself listed; do it and log it.new— valuable but unrequested; do NOT do it, queue it.
Log every
adjacentandnewitem toSCOPE.mdwith an estimate in minutes and the criterion that spawned it.Before starting a
newitem, check the time budget: once elapsed exceeds 80% of the box, stop adding work and start reporting.Surface
newitems to the user as a labelled list at the next report, never as a finished fait accompli.When a discovered bug blocks a criterion, it is
adjacentand belongs in the log; when it does not, it isnewand belongs in the queue.Count deltas at the end:
grep -cE "^(adjacent|new)" SCOPE.md, and reconcile against what you actually did.
Pitfalls
- Refactoring en route feels free but invalidates prior test runs and re-opens finished work.
- "It's only two lines" ignores the review, test, and rollback cost those two lines carry.
- A
newitem completed silently makes the deliverable differ from what was agreed. - Renaming cleanups (typos, formatting) still produce a diff someone must review; log them.
- An
adjacentitem that keeps growing into a subsystem is really anewproject; reclassify.
Verification
grep -cE "^(in-scope|adjacent|new)" SCOPE.md; grep '^new' SCOPE.md # queue shown, none executed
Report the added-item count and every deferred item with its estimate, one line each.