AI prototype review

Share an AI prototype. Get feedback.

Host a self-contained HTML prototype at an unguessable link for 14 days. Send it to whoever you want; they pin comments right on the page — no signup, no accounts. You (and Claude Code) pull a structured synthesis — with the screenshot each reviewer saw — and iterate.

Open source · deleted after 14 days · no signup · no accounts

Anyone with the link can view and comment. There's no allowlist and no gate — that's the trade for how little there is to set up. Don't put anything sensitive in a ProtoPeek prototype.

Install · one line in your terminal

Add the skills to your coding agent:

npx skills@latest add Brightwing-Systems-LLC/protopeek -g --skill '*'
/proto-up /proto-status /proto-feedback
/proto-list /proto-delete
  • All five inspectable skills at once, straight from the open-source repo — installed for your user, not one project
  • Claude Code, Codex, Cursor — any Agent-Skills-compatible tool
  • No signup, no password — the first /proto-up mints an anonymous token after asking you
Other ways to install
claude plugin marketplace add Brightwing-Systems-LLC/protopeek
claude plugin install protopeek@protopeek

Prefer manual? Your agent can follow https://protopeek.dev/agent.md — it asks before every consequential step.

◉ Deleted on a deadline

14 days after you upload it, the prototype is destroyed — link, every version, all feedback, all screenshots. Nothing extends that, so there is nothing to revoke and nothing to clean up.

⌖ Pin-anchored notes

Reviewers annotate elements directly. Author stamped server-side.

↻ Agent-shaped

Pull a synthesis — themes, conflicts, a change list — not a comment dump.

▣ Sees what they saw

Every pin ships the screenshot, the reviewer's box highlighted. Your agent reads the pixels, not just the selector.

The whole loop

From prototype to synthesized feedback

A real run: a Tempo pricing page, reviewed by three people, pulled back and synthesized by the agent.

A reviewer drawing a box on the Team plan's CTA to leave an element-anchored comment
Reviewers pin comments right on the page — every note anchored to its element and stamped to its author.
01

Install — one line, no signup

One command in your terminal installs the five skills from the open-source repo. Nothing else to configure — the first /proto-up mints an anonymous token (no account, no password) after asking you.

npx skills@latest add Brightwing-Systems-LLC/protopeek -g --skill '*'
proto-up · proto-status · proto-feedback proto-list · proto-delete installed for Claude Code (works in Codex & Cursor too) Ready — try /proto-up ./prototype.html in your agent
02

Upload a prototype

Generate a self-contained HTML file, then one command uploads it and returns an unguessable, time-boxed link.

/proto-up ./tempo-pricing.html
https://reviews.protopeek.dev/p/d2a6…39b8 (v1) anyone with this link can view and comment this prototype will expire and be purged at 14:32 on Tue 11 Aug 2026
The Tempo pricing-page prototype
The self-contained prototype you upload.
03

Reviewers see it & pin comments

Send the link to anyone. They type an email once — a name tag, not a password — and draw a box on anything to leave a comment. It anchors to that element.

A reviewer drawing a box on the Choose Team button to leave a comment
Pinning a comment on the Team plan's CTA.
04

Pull the feedback

When comments land, your agent pulls an agent-shaped payload — every note with its author and the element it targets.

/proto-feedback d2a6…39b8
3 reviewers · 3 comments · 3 new #47 sam @ #plan-team "Pro and Team both carry filled buttons…" #48 jordan @ #plan-team "Disagree — Team's dark button is how…" #49 taylor @ #toggle "Monthly/Annual toggle is too quiet…"
The collected feedback panel showing all comments
Everything reviewers left, in one place.
05

Your agent sees what they saw

Every pin carries a screenshot with the reviewer's box highlighted in orange. /proto-feedback downloads them and your agent looks at the pixels — so a vague note like "this competes" is settled by the picture, not guessed from a CSS selector.

A saved reviewer screenshot with an orange box highlighting the Choose Team button the reviewer drew around
The screenshot your agent reads — the reviewer's box, highlighted.
06

Discuss & iterate with your agent

Your agent doesn't dump comments — it synthesizes: themes, who-said-what, conflicts, and a concrete change list mapped to the page. You decide; it edits and re-shares as v2.

your agent · synthesis

3 reviewers, 3 notes on v1. The core tension is CTA hierarchy — which tier are you steering people to?

Sam wants only the recommended tier to carry a filled button — Pro and Team both being loud makes them compete. Jordan pushes back that Team's dark button is how enterprise buyers self-select, and would keep it. That's a real product question: is the page funneling everyone to Pro, or letting big teams find Team on their own? Taylor flags a separate fix above.

Proposed v2:

  • Commit to one primary CTA — keep Pro solid, demote Team's button to an outline (or make Team the dark one only if it's the real target)
  • Make the Monthly/Annual toggle louder — surface the −20% as a visible savings pill
  • Point the "Most popular" badge at whichever tier the CTA actually favors, so flag and button agree

Want me to make these edits and re-share as v2?

npx skills@latest add Brightwing-Systems-LLC/protopeek -g --skill '*'

One line in your terminal, then run the loop from your agent. Read the source or the agent setup page first if you like.

Need to ask a specific person, or keep the history? That's HumanGated.