detecting-scope-creep

Detecting scope-creep before it happens

Scope-creep is the most common way to blow a phase deadline. Each "just one more little thing" costs time, attention, context, and risks the phase gate.

When this skill activates

Auto-activates when the context involves phrases like:

• "Since I'm already here, I'll also..."
• "While I'm at it, I can add..."
• "It also makes sense to..."
• "I'll just quickly fix..."
• "We also need..."
• The proposal mentions a capability not listed in the active feature spec

Procedure (3 steps)

Step 1 — Identify the active feature + read the scope

1. Consult CLAUDE.md section 4 — what is the active anchor feature?
2. Read the feature's docs/features/F<N>-*.md.
3. Mentally list the IN scope and the OUT scope.

Step 2 — Classify the proposal

Which category does the newly proposed thing belong to?

• (A) Within the IN scope of the active feature — OK. You may proceed.
• (B) Within the explicit OUT scope — STOP. Don't do it now.
• (C) Not mentioned in IN or OUT — gray area. Decide consciously.

Step 3 — According to the classification

(A) It's in the IN

Proceed. Nothing to do.

(B) It's in the explicit OUT

Decline clearly. Suggested response:

"This is marked as OUT scope in docs/features/F<N>-*.md. Reason: . If you've changed your mind, we can:

1. Defer it to the next phase (F<N+1>).
2. Open a new feature spec via /feature-spec if it has become a priority.
3. Consciously update the current feature (move it from OUT to IN with justification).

For now, I'll follow the original scope."

(C) Gray area

Present the user with 3 conscious-decision options:

1. Add it to IN now — if it's critical for the current phase. Update the feature spec.
2. Defer to the next phase — record it via /todo or in the project backlog.
3. Quick spike (/spike) — if the question is "is it even worth it?".

DO NOT code without an explicit decision.

Examples

✅ Good

Eric: "Since I'm going to implement the anchor feature, I might as well also add detailed metrics (capability Z)."

Skill activates: "I checked docs/features/F1-<slug>.md § Scope. Instrumentation is indeed listed as a capability. BUT the specific capability Z is not mentioned — F1 only covers the basics. Status: gray area.

Options:

1. Keep only the basics in Phase 0 (aligned with the plan).
2. Add capability Z now (update the F1 spec first).

I recommend option 1 — capability Z is planned in a later-phase feature (F7-<slug>.md). Adding it now turns into scope-creep of a future phase pulled into Phase 0."

❌ Bad

Eric: "While I'm at it, I'll do basic auth at the edge."

Skill doesn't activate or activates but is ignored: implementation moves forward.

Result: Phase 0 (which has "no auth" in the OUT scope) gains untested, unreviewed auth code, and Phase 2 will inherit the debt. The Phase 0 gate slips.

Special case: "preparatory" refactor

Typical phrase: "I'll go ahead and prepare the code for the multi-tenant support coming in Phase 2."

Generally NO. It's part of YAGNI discipline:

• Building an abstraction now for the "future" produces unvalidated code.
• By the time Phase 2 arrives, the requirements may have changed.
• The @architect would probably reject this in review.

Exception: if the cost of doing it now is trivial AND the cost of doing it later is high (e.g., a column name). Then it's OK, but DECLARE it.

Anti-patterns of this skill

• ❌ Accepting a gray area as "when in doubt, do it" — force a decision.
• ❌ Confusing a "quality refactor" (within the code you're already touching) with scope-creep (a new capability).
• ❌ Not consulting the active feature spec before classifying.