Comment to Fix Start now

For teams reviewing AI-built pages, decks, and prototypes

Turn visual feedback into code-ready fixes.

Comment to Fix lets you click any element in an HTML preview, leave a precise note, and hand your coding agent the exact context it needs to make the change.

Install the skill Watch the loop
No design handoff doc No vague screenshots No lost context
Make this CTA more specific

Use the install action and show the copied command nearby.

Agent fix queued

Element selector, text context, viewport, and comment captured.

Purpose-built for agent feedback loops Review static HTML, annotate exact elements, and keep every fix traceable.
Works where agents already work Codex, Claude Code, Cursor, and any workflow that can edit local files.
Designed for fewer review rounds Every comment includes the page context an agent usually has to guess.

See the actual loop

Point at the problem. The agent gets the fix brief.

Instead of writing "the hero feels off" in chat, reviewers annotate the exact headline, button, card, or spacing issue. The agent receives the comment plus structured element context, edits the file, and waits for the next note.

The workflow

A landing page review loop built for precision.

  1. 01 Open a local HTML preview

    Launch the overlay on the page, slide deck, proposal, or prototype your agent generated.

  2. 02 Comment on the exact element

    Click, select, or target the thing that needs work. The note stays attached to the page context.

  3. 03 Send a better task to the agent

    The agent receives the selector, surrounding copy, viewport, screenshot context, and your instruction.

  4. 04 Review, approve, or repeat

    Markers show status as the agent edits. Keep commenting until the page is ready.

Why teams switch

Most feedback is too vague for an agent. This makes it actionable.

Screenshots and chat notes drop the details agents need: which node, which state, which nearby text, and which file changed. Comment to Fix packages those details into the review loop.

Before

  • "Move this up a bit" in chat
  • Annotated screenshots detached from code
  • Repeated explanations after every revision

After

  • Element-specific comments
  • Agent-ready context captured automatically
  • A repeatable comment, fix, review loop

Built for local-first review.

Run it against files in your workspace. Keep the page, comments, and agent edits inside the same development loop.

  • Works with static HTML output
  • Captures comment state for follow-up
  • Supports repeat review sessions
  • Pairs with existing coding agents

Try it from zero

Paste these into Claude or Codex, one after the other.

If you already have an HTML page, skip step 1. If you just want to try the workflow, start by asking your agent to make one.

  1. Step 1 · Get an HTML file

    If you have no HTML yet, paste this into Claude or Codex first:

    Generate HTML-based slides for my startup idea. Ideate something strong and save it as an HTML file.
  2. Step 2 · Install the skill

    Paste this into the same Claude or Codex chat:

    npx skills add ralfboltshauser/comment-to-fix
  3. Step 3 · Restart your agent

    Close and reopen Claude or Codex so it can see the new skill.

  4. Step 4 · Run Comment to Fix

    Paste this into Claude or Codex. It will open your HTML and start watching for comments:

    /comment-to-fix for my HTML slides

    Replace "HTML slides" with "landing page", "proposal", or whatever HTML file you want to review.

Simple version: Paste step 2, restart, then paste step 4. Comment to Fix will go to work.
View on GitHub