What does this skill do?

The Skill Review Canvas opens plans and HTML artifacts in a browser canvas where the user annotates specific elements, chats, and issues a verdict of “Approve” or “Request Changes.” The agent waits for a single CLI call that returns all the feedback in JSON format, streamlining the approval cycle without the need to describe the changes in words.

Approval of Plans
After creating a plan in .claude/plans, open it in your browser and get a structured assessment before implementing it.
Anchored annotations
The user specifies the exact DOM element they want to change; the skill returns the selector and a snippet of the specified text.
Embedded Chat
Unrestricted messaging within the canvas with direct responses from the agent, without having to return to the terminal.
Change Loop
Each "request-changes" resets the cycle: edit the artifact, respond, and wait again for the next verdict.

Usage examples

📋 Open plan for review
Open the notification plan in your browser so you can review it before implementing it.
🖍️ Add annotations
I've noted that Phase 2 needs to be divided into two parts—review the plan and continue the loop.
🎨 Mermaid Diagram
Draw the migration flow as a Mermaid diagram within the plan so you can review it on the canvas.
✅ Approval verdict
Now that the payment feature plan has been approved, implementation begins.

Features

Canvas in the browser Open local .md or .html files in an interactive page using `ecc-plan-canvas open`.
Annotations Anchored to the DOM Each entry includes a CSS selector and a code snippet for the exact element selected by the user.
Structured verdicts Returns "approve" or "request-changes" in a JSON format ready for the agent to decide the next step.
Two-way chat Reply within the canvas using --reply without interrupting the next feedback session.
Mermaid Rendering Mermaid blocks are drawn as visual diagrams that the user can point to directly.

Frequently asked questions

Yes, the skill uses the ecc-plan-canvas CLI to open the canvas, wait for feedback, and close the browser session.
No. The canvas only supports local .md or .html files. It is not used to review code diffs or run web applications.
Nothing is lost: the appended feedback is returned when you run `ecc-plan-canvas await` again on the same file.
If the user closes the canvas, `await` returns the status "ended" with `endedBy` set to "user." Do not reopen the canvas unless you explicitly request it.
Review Canvas — Visual Approval of Plans with Claude AI

¿Prefieres escuchar el contenido? Genera la narración de audio con un clic.