Sno Station
Sno Station is your agents' workstation: open-source software that turns the AI agents you already run into one squad on your own machine. Two heads are better than one, and they get better with every
Senior Product Designer working on B2B SaaS, data-heavy platforms and AI workflows. Case studies in clinical diagnostics, lab management and veterinary software.
PluginsPython · JavaScriptGitHub @batchencba-beepOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
No updates detected yet. GitHub releases will appear here.
Owners of public repositories under a personal GitHub account can sign in to verify ownership and manage the introduction themselves. No administrator approval is needed.
Sign in, generate a verification file, and commit it to this repository’s default branch within 30 minutes. Files on other branches or forks do not count. Then check the file below to manage the project without administrator approval. Keep the file to retain your access.
If needed, you will go to sign-in. After signing in, return here and press your chosen verification button again.
Run /nomorevibe verify in your project folder to verify ownership and manage the introduction yourself.
Verify in one minuteAsk about other verification options
A Claude Code plugin that turns design review findings into a decision board: Today vs Proposed, Approve / Reject / Discuss, copy back as a prompt. Beta.
Taken from the repository README
Loading the full README. The saved excerpt is shown below.
Beta. It works, I use it daily, and I'd love to hear where it breaks for you. See Feedback.
A Claude Code skill that turns a pile of proposals into a page you decide on.
When an agent reviews your design, it usually answers with a long list you have to read, picture in your head, and reply to in text. With this skill it opens a page instead: one card per proposal, Today next to Proposed, and three buttons. You look, click, add a note where you disagree, and one button copies every decision back to Claude as a prompt.Read the full README ↗Beta. It works, I use it daily, and I'd love to hear where it breaks for you. See Feedback.
A Claude Code skill that turns a pile of proposals into a page you decide on.
When an agent reviews your design, it usually answers with a long list you have to read, picture in your head, and reply to in text. With this skill it opens a page instead: one card per proposal, Today next to Proposed, and three buttons. You look, click, add a note where you disagree, and one button copies every decision back to Claude as a prompt.
The idea comes from Kyle Zantos, who showed his review interface for agent proposals on Dive Club (August 2026): gather the proposals, show each one next to how things look today, approve, deny or discuss it, and copy every decision back into the session. This is an open implementation of that approach.
Claude never writes the page. It writes one short findings.json, and a small Python script turns it into HTML. A review round costs a few hundred tokens of data, not a hand-built page every time.
Claude Code (recommended). Two commands, then start a new session:
claude plugin marketplace add batchencba-beep/decision-review
claude plugin install decision-review@batchenba
Updates later: claude plugin update decision-review@batchenba.
Claude desktop or web app. Download this repo as a ZIP and upload it in Customize → Plugins → Upload plugin. The decision page works there too; the screenshot script needs Node and Chrome, so in the apps expect text and swatches rather than screenshots.
By hand. Copy skills/decision-review into ~/.claude/skills/ (every project) or into a project's .claude/skills/ (that project only).
Ask for it directly:
Review the homepage and give me a decision page.
Or just run a review: the skill tells Claude to use the page whenever more than two or three changes are waiting on you. When you're done deciding, click Copy all decisions as a prompt and paste it back into the chat. Approved items get applied, rejected ones stay as they are, notes override the proposal, and Discuss items get an answer before anything changes.
findings.json, builds the page and opens it.Your clicks are saved in the browser, so you can close the page and come back. A new review round with different items starts clean.
To try it by hand, without touching the skill folder:
cp -r skills/decision-review/example /tmp/decision-review-example
cd /tmp/decision-review-example
python3 /path/to/decision-review/skills/decision-review/build.py findings.json decisions.html
open decisions.html
It isn't only for web pages. Point Claude at a Figma file or a deck and it reviews the frames, crops what matters, and shows the proposal next to it. With a Figma connector it takes the frames itself; without one, export them and it works from the images.
Most real screens sit behind a sign-in. The first time Claude needs one, it opens a normal Chrome window on the app's sign-in page. You sign in yourself and close the window, and from then on screenshots are taken as you, signed in.
That window uses a separate Chrome profile that belongs only to this tool, stored in ~/.decision-review/. Claude never sees or types your password, and your own Chrome profile, cookies and passwords are never touched. To delete the saved session: node skills/decision-review/shoot.mjs --logout, or delete that folder.
| File | What it does |
|---|---|
SKILL.md | Instructions for Claude: when to use it, the JSON schema, how to write the cards |
build.py | Turns findings.json into the page, and explains what's wrong if the JSON isn't right |
shoot.mjs | Screenshots one element of a page, as it is or with proposed CSS applied, signed in if needed |
example/ | A ready example to build and open |
Language. The page comes in the language you talk to Claude in. English and Hebrew are built in (Hebrew renders right to left, and mixed Hebrew/English text keeps its direction). Want the other one? Just say "decision pages in English" or "בעברית". Any button label can be changed with "labels" in the JSON.
This is a beta. If something breaks, reads wrong, or you wish a card worked differently, open an issue or message me on LinkedIn. A screenshot of the page and the findings.json that produced it help the most.
MIT