What problem does it solve? Getting plan approval usually means pasting a plan into chat and waiting for a typed yes, which loses precision when feedback refers to specific sections. This Skill serves local plan or HTML artifacts in a browser canvas where the reviewer annotates exact elements, chats, and returns an approve or request-changes verdict as structured JSON. ## Core Features & Use Cases - Browser-based review loop: Serves local .md and .html artifacts on a loopback server with an annotation layer, so reviewers point at the exact element they mean. - Structured feedback via CLI: The await command blocks until the human responds and returns JSON containing annotations, chat messages, and an approve or request-changes verdict. - Mermaid diagram rendering: Renders fenced mermaid blocks as themed diagrams in the canvas, degrading to source text when the CDN is unavailable. - Use Case: After /plan writes .claude/plans/notifications.plan.md, open it in the canvas, block on await, and begin implementation the moment the reviewer clicks Approve plan. ## Quick Start Open my plan file in the browser canvas and wait for my approval or annotated feedback before you start implementing.