cursor-task-scoping

What problem does it solve?

Scopes Cursor Agent work into one clear step per response and clear outcomes. Use when planning work, starting a complex task, or when the user wants smaller steps.

Core Features & Use Cases

• One step per response (default): Default: one step per response for large or risky work. Medium tasks: up to two tightly-related steps when safe and reviewable. Small tasks: multiple tiny actions in one response when bounded and safe.
• Atomic task discipline: One outcome per task — Avoid "and" in task titles; split multi-verb tasks. Verb-noun naming — e.g. "validate-input", "save-settings". Testable in isolation — No circular dependencies between steps.
• Step subdivision: If a step is large, split into labeled substeps (Step 4a, 4b, 4c). If context or time is at risk, split into step parts and complete with validation and rollback per part.
• Output contract: 1. Questions — Only as many as needed. 2. Prerequisites — Auto-run safe in-repo prereqs when possible; log actions. 3. Next step — Usually one; at most two when tightly related and low risk.

Quick Start

Provide a plan that breaks the user's task into a single, focused step per reply with clear outcomes.