| name | scope-check |
| description | Assess whether a proposed change fits the current work context or represents scope creep |
| argument-hint | ["description of proposed change"] |
Scope Check Assessment
Evaluate whether the following proposed change aligns with the current work context.
Proposed Change: $ARGUMENTS
Assessment Steps
-
Read recent CHANGELOG entries — Check CHANGELOG.md for the last 2-3 dated sections to understand recent work themes and feature areas.
-
Review recent commit history — Run git log --oneline -15 to see the pattern of recent work. Note which feature areas, file paths, and types of changes dominate.
-
Check in-progress work — Run git diff --name-only and git diff --staged --name-only to see what is actively being modified right now.
-
Identify the current work theme — Based on the above, summarize:
- What feature area(s) are currently active?
- What kind of work is being done (new feature, bugfix, refactor, data entry)?
- Are there open threads of work that are incomplete?
-
Assess the proposed change — Evaluate $ARGUMENTS against the current context:
Natural Extension (GREEN): The change...
- Directly relates to files currently being modified
- Fixes a bug discovered during current work
- Completes an incomplete feature from recent commits
- Is a small improvement in the same feature area
Adjacent Work (YELLOW): The change...
- Touches a related but different feature area
- Could be deferred without impacting current work
- Requires context-switching but shares some code paths
Scope Creep (RED): The change...
- Is in a completely unrelated feature area
- Requires significant new infrastructure
- Would interrupt incomplete in-progress work
- Introduces new dependencies or architectural patterns
-
Provide recommendation — State the assessment (GREEN/YELLOW/RED) with reasoning. If YELLOW or RED, suggest:
- Whether to defer (and what to finish first)
- Whether to create a GitHub issue to track it instead
- A minimal version that could be done now vs. later
Important
This is a READ-ONLY assessment. Do not make any code changes. Only analyze and advise.
Scope-reduction guidance (Pick #18 from ecosystem-review adoption)
Enforcement level: GUIDANCE. This is a discipline rule the skill enforces by following its own instructions, not a deterministic gate. There is no hook, plan-vs-delivery diff, or other mechanical verifier. Codex review #29 flagged this honestly: a prose-only rule depends on Claude reading and following it. Treat it as a routing nudge during scope reasoning, not a guarantee.
When the proposed change exceeds the current task's scope, never silently drop part of it. The recurring failure mode: agent decides "I'll skip Y to keep this manageable" without flagging, and Y is actually load-bearing for the user. The user discovers Y was dropped two commits later when something downstream breaks.
The required pattern:
- List every concern in the proposed scope. Don't pre-edit.
- For each, classify: in-scope-now, fits-as-follow-up, or genuinely orthogonal-and-deferred.
- If anything would be silently omitted from the in-scope-now bucket, STOP and present the proposed split:
This change has scope-creep. Two ways to handle it:
(A) Phase split — ship now / defer later:
Now (this commit): X1, X2, X3
Follow-up: Y1, Y2 (open issue + estimate)
(B) Wider scope — do everything in this commit (estimate: <effort>)
Which? (Default: A — split.)
- The third option — "drop Y silently" — is not on the menu. If Y is genuinely garbage, that's a separate "delete Y from the spec" decision that the user makes explicitly.
Source: gsd-build/get-shit-done's planner-source-audit pattern. Catches the class of bug where "I trimmed the scope to fit" silently discards user-stated requirements.