evaluate-planlisted

# Evaluate Plan Systematically verify that an AI-generated implementation plan fully covers its source requirements. **Context**: First-pass plans silently drop ~20-40% of explicit requirements. A structured coverage check recovers most gaps. This skill forces a coverage-first second pass. ## Phase 1: Identify the Source of Truth Locate the document(s) the plan was built from: - A PRD, spec, feature brief, or user story set - An intent document or voice transcript - The original user request in this conversation - A file path provided as argument (e.g., `evaluate plan against @prd.md`) If multiple sources exist, treat them as a combined requirement set. **If no explicit document exists**: reconstruct the requirement set from the user's messages in the current conversation. Quote the original requests. > **Rule**: The source document is the authority. If something is in the source but not the > plan, it is a gap — regardless of whether the plan's approach seems reasonable without it. ## Phase 2: Extract Requirements Inventory Read the source and extract every discrete requirement into a flat checklist: | Category | What to Extract | |----------|----------------| | **Functional features** | Every capability, behavior, or feature described | | **Interaction behaviors** | Click, hover, drag, dismiss, keyboard, gesture behaviors | | **UX / emotional intent** | How it should *feel*, experience goals, tone | | **Visual requirements** | Layout, styling, animations, respons