Skip to content

Repository files navigation

Visual Stack

Visual Stack

Stop prompting. Start pointing.

Visual Stack adds a Figma-like feedback layer to AI coding agents.

Create a new screen, or open an existing app. Click anywhere and leave highly professional feedback such as:

"claude its 3am just align the buttons"

"why is everything a card. who hurt you"

"pls undo the series b dashboard energy we have 4 users"

"the padding is now between you and god"

Your agent applies the feedback and updates the design in the same place, hopefully without adding another gradient.

Point. Comment. Iterate. No more jumping between chat and the UI. Point. Comment. Iterate. No more jumping between chat and the UI.

Get started

Claude Code

Install, in Claude Code:

/plugin marketplace add Cavalry-Collective/visual-stack
/plugin install vstack@cavalry-collective

Then run:

/vstack:review Wireframe a desktop personal task manager with minimal aesthetics.

Codex

Install, in your terminal:

codex plugin marketplace add Cavalry-Collective/visual-stack
codex plugin add vstack@cavalry-collective

Then run:

$vstack:review Wireframe a desktop personal task manager with minimal aesthetics.

What you can do

  • Work in a familiar, Figma-like interface.
  • Click any element and leave feedback exactly where the problem is hiding.
  • Drag a thing to where it belongs, or strike out what should go — no note required.
  • Stay in the workspace as your agent publishes each update.
  • Preview desktop, tablet, and mobile layouts before production does it for you.
  • Compare revisions and identify the exact moment things went wrong.
  • Replace “you know what I mean” with actual context.

Why Visual Stack?

Design feedback is spatial. Chat is linear. Humans are tired.

As feedback becomes more visual, chat becomes the bottleneck. Context gets buried in long conversations. Screenshots go stale. Describing which element you mean becomes a small detective novel.

Visual Stack keeps every comment attached to the element, route, and version it refers to. Your agent receives the feedback with the visual context intact.

No archaeology through 200 messages. No screenshot named final-final-v2-actually-final.png. No arguing about which blue.

Requirements

  • A current version of Claude Code or Codex
  • A supported Node.js LTS release
  • Git
  • A local web browser
  • At least one strong opinion about border radius

Technical Details

Live Link

Each workspace is linked to one agent session. The link holds while that session is active, its heartbeat is less than 15 seconds old, and every submitted review round has been claimed.

The workspace page in a browser tab talks over http and SSE to the review server on 127.0.0.1. The server reads and writes a store on disk holding the state, the versions, the comments, the rounds, and the files that carry the link. The agent session watches and writes the same store.

Review Lifecycle

Your comments are submitted as one review round. The agent claims the round and reads its brief, asking for clarification when a comment is unclear. Comments sent while the round is in progress join it. Publishing is blocked until every comment has been applied, answered, or dismissed, and the published version appears in the same workspace.

Contribute

Visual Stack is open source and under active development. Expect rough edges, breaking changes, and occasional moments of character development.

Feedback and contributions are welcome.


Built by DeyangChan, presumably after one too many rounds of screenshot-based feedback. Released by Cavalry Collective.

Licensed under the MIT License.

About

A visual interface for building software with Claude Code/Codex.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

3 stars

Watchers

1 watching

Forks

Releases

Packages

Used by

Contributors

Languages