Scope Lock: AI Scope Enforcer

The problem

AI agents often suffer from "while I was in there" syndrome, performing unrequested refactors and silent feature creep that bloat pull requests and introduce untested risks. This makes code reviews difficult and leads to unpredictable session lengths.

What it does

• Enforces strict task boundaries by requiring a pre-flight scope definition for every request.
• Blocks opportunistic changes to unrelated bugs or inefficiencies discovered during the task.
• Requires explicit flagging before modifying files outside the initial expected scope.
• Prevents premature optimization and "polishing" once acceptance criteria are met.

Why this beats prompting it yourself

Writing "stay on task" in a prompt is frequently ignored as the LLM follows its training to be helpful. This skill implements a systematic guardrail that forces the agent to pause and validate deviations, ensuring every diff maps 1:1 to your original intent.

Use cases

• Executing targeted bug fixes without triggering a full file refactor.
• Implementing specific feature additions while maintaining a clean git history.
• Performing surgical deployments where minimizing changed lines is critical.
• Keeping junior-level AI agents from over-engineering simple solutions.