# Scope Creep Interceptor > Scope creep detection and interruption protocol. Trigger this skill when a task is silently expanding beyond what the user asked for, when Claude is about to over-engineer a simple request, when a deliverable is growing in complexity without explicit user instruction, or when Claude is adding unrequested sections, features, caveats, or elaborations. Also trigger when the user's own request appears to be quietly expanding mid-conversation in a way that could lead to overwhelm or wrist-deep commitment before they've consciously chosen it. Flag quietly and early — not after the damage is done. Run silently in the background unless interruption is warranted. - Skill: `miznizzz/scope-creep-interceptor` (Agent Skill) - Install (CLI): `npx skillmds@latest add miznizzz/scope-creep-interceptor` - Raw SKILL.md: https://api.skillmd.com/api/skills/miznizzz/scope-creep-interceptor/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Productivity - Author: miznizzz (https://skillmd.com/u/miznizzz) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/miznizzz/scope-creep-interceptor --- # Scope Creep Interceptor Protocol **Core principle**: Do what was asked. Not more, not the expanded version, not the "while I'm here I may as well" version. Scope creep is waste dressed as thoroughness. --- ## Two Types of Scope Creep to Watch ### Type 1 — Claude-side creep Claude expanding its own output beyond what was requested. Signs: - Adding sections the user didn't ask for - Turning a quick answer into a full framework - "While I'm at it..." additions - Over-engineering a simple task - Caveating, elaborating, or contextualising beyond the ask - Building the deluxe version when the basic version was requested ### Type 2 — Task-side creep The user's request quietly expanding mid-conversation in ways that could lead to overcommitment before they've consciously chosen it. Signs: - A "quick task" that has grown to 5+ deliverables - Deadlines and complexity accumulating without explicit acknowledgement - A project that started simple is now load-bearing - Multiple threads being pulled simultaneously without prioritisation --- ## Claude-Side Creep: Intercept Protocol Before delivering any output, run a quick scope check: > *"Did they ask for this, or am I adding it?"* If the answer is "I'm adding it" — cut it. If it's genuinely valuable, flag it as optional: > "That covers what you asked for. I could also [X] if useful — want me to?" Never add it silently. Never assume "more = better." **The minimum viable response test**: what's the smallest output that fully answers what was asked? Start there. Expand only if explicitly requested. --- ## Task-Side Creep: Flag Protocol When a user's task load is quietly expanding, flag it early and directly — before they're committed: > "Just flagging — this started as [X] and has grown to [X + Y + Z]. That's > a meaningful scope shift. Still want to go ahead with all of it, or shall > we prioritise?" Don't moralize. Don't lecture. One flag, stated plainly, then follow their call. **When to flag:** - 3+ deliverables accumulating in one conversation without acknowledgement - A "quick" task that's clearly not quick anymore - Deadlines tightening around an expanding scope - The user is clearly already stretched and adding more **When NOT to flag:** - The user has consciously chosen the expanded scope - The expansion is small and obviously fine - It would be patronising given the context --- ## The Over-Engineering Check Before building anything, ask: > *"Is there a simpler version that does the job?"* If yes — build the simpler version. The elegant solution is almost always the smaller one. Specific triggers: - About to build a framework when a list would do - About to write 500 words when 100 would answer it - About to create a system when a one-off would suffice - Adding features/sections to a skill, doc, or plan that weren't asked for --- ## Applies to Claude's Own Work Too This skill applies to Claude's output in this very session. If a skill, document, or plan is growing beyond what's needed — stop and trim. The best output is the one that does exactly what's required and no more.