product-review— agent skill

Preamble (run first)

# Suiperpower telemetry. Routes through the CLI so projects.json + Convex stay in sync.
# Silent on failure so the skill never blocks. Opt out: SUIPERPOWER_TELEMETRY=off.
#
# AGENT NOTE: when this skill finishes, run the matching completion command:
#   suiperpower track product-review build completed
# Or use "failed" / "aborted" if it ended that way.
command -v suiperpower >/dev/null 2>&1 && suiperpower track product-review build started >/dev/null 2>&1 &
true

If TEL_PROMPTED is no, before doing real work, ask the user:

Help suiperpower get better. We track which skills get used and how long they take. No code, no file paths, no PII. Change anytime in ~/.suiperpower/config.json.

>

A) Sure, anonymous B) No thanks

Write the answer to ~/.suiperpower/config.json telemetryTier field and create ~/.suiperpower/.telemetry-prompted. Then continue.

What this skill does

Walks the product as a first-time user would. Catalogues issues across five dimensions: first-paint experience, onboarding friction, empty / error / loading states, mobile experience, and time-to-first-value. Produces a roadmap of UX fixes ranked by impact divided by effort.

Less brutal than roast-my-product, more concrete. Where roast is about whether the project survives a hostile reading, product-review is about whether the user gets to the value with as little drag as possible.

When to use it

• Before submitting a hackathon project, alongside or after roast-my-product.
• After a build sprint, when the user has shipped enough surface area to have a real walkthrough.
• Before a marketing push, when first-time-user friction will determine conversion.

When NOT to use it

• Pre-MVP, before there are screens to review.
• For internal tools the team uses every day. They will not feel the friction a first-time user feels.
• For pure protocol projects with no app surface. Route to review-move for code review or ottersec-prep for security.

If you activated this and the user actually wants something else, consult skills/SKILL_ROUTER.md and hand off.

Inputs

• .suiperpower/idea-context.md for the chosen idea and target user.
• A live URL for the product, or screenshots / a recorded walkthrough.
• The user's claimed "happy path" for a first-time user (the steps they expect a new user to take).
• If a wallet flow exists: the user's intended wallet (Slush, Suiet, Phantom on Sui, etc.) and whether zkLogin is offered.

Outputs

A .suiperpower/product-review.md with findings across five dimensions and a prioritized roadmap:

## Product review, <timestamp>

### Inputs reviewed
- live URL: <link or "not provided">
- claimed happy path: <one paragraph>
- target user: <one sentence>

### First-paint experience
- finding 1: <observation>, <impact>
- finding 2: ...

### Onboarding friction
- finding 1: ...
- finding 2: ...

### Empty / error / loading states
- finding 1: ...
- finding 2: ...

### Mobile experience
- finding 1: ...
- finding 2: ...

### Time-to-first-value
- measured TTFV: <seconds, from landing to first concrete value>
- target TTFV: <seconds, based on category>
- finding 1: ...

### Roadmap (prioritized by impact / effort)
- P0 (high impact, low effort): <numbered list, ship before next demo>
- P1 (high impact, medium effort): <numbered list, ship within a week>
- P2 (medium impact, low effort): <numbered list, ship when time allows>
- defer: <findings that are not worth the effort right now>

Workflow

1. Read context and inputs

• Pull idea, target user, claimed happy path. If no happy path is provided, demand one.
• Open the live URL in a fresh browser session, no cached state, no logged-in account. If the user has not provided a URL, request screenshots of every page on the path.

1. First-paint experience

• What does a first-time user see in the first 5 seconds?
• Is the value clear from the headline alone? Is the CTA visible above the fold? Is the typography load free of font-flash?
• Note any layout shift, missing alt text, hero copy that does not name the user or the outcome.

1. Onboarding friction

• From "first paint" to "first concrete value", count the steps.
• For each step, note: is it necessary, is the copy clear, is the input format obvious, are there error fallbacks?
• If the path includes wallet connect, count it as one step. If it includes zkLogin, note the OAuth provider load time.
• Reject "we use industry-standard wallet flow" as defense; the friction is the friction.

1. Empty / error / loading states

• Visit each state deliberately:
• Empty: a brand new user with no data. Is the empty state useful or a blank panel?
• Error: cause an error (network off, invalid input, expired session). Does the message explain what to do?
• Loading: throttle the network. Are skeletons shown? Does the UI block, or does it stay responsive?
• Note any state that returns a blank screen, an unstyled error, or an infinite spinner.

1. Mobile experience

• Open on a real phone (or 375 px viewport). Does the layout reflow? Are tap targets at least 44 px? Does the wallet connection complete on mobile, including zkLogin redirects?
• Note any horizontal scroll, fixed-positioned elements that block content, or buttons that fail off-screen.

1. Time-to-first-value

• Measure: from landing page load to the user receiving the first concrete output (a transaction confirmed, a piece of content rendered, an answer returned).
• Compare to the category benchmark in references/ttfv-benchmarks.md.
• If TTFV exceeds the benchmark by 2x, this is a P0 finding regardless of cause.

1. Score and prioritize

• Apply the impact / effort matrix from references/impact-effort-matrix.md.
• P0: high impact, low effort. Ship before next demo.
• P1: high impact, medium effort. Within a week.
• P2: medium impact, low effort. When time allows.
• Defer: low impact or high effort with no payoff.

1. Writeback

• Write .suiperpower/product-review.md. Surface P0 to the user verbally.

Quality gate (anti-slop)

Before reporting done:

• Did the review actually walk the live product, not just the deck?
• Was the claimed happy path stress-tested with a fresh browser session, no cached state?
• Are findings specific (page, element, observation), not "the UX feels heavy"?
• Is TTFV a measured number, not a guess?
• Does the roadmap rank by impact / effort, not by what the user wants to fix?
• Did the writeback happen?

If any answer is no, the skill keeps working.

References

On-demand references (load when relevant to the user's question):

• references/impact-effort-matrix.md: How to assign impact and effort scores per finding.
• references/ttfv-benchmarks.md: Time-to-first-value targets by product category (DEX, social, indexer, NFT marketplace).

Use in your agent

• Claude Code: claude "/suiper:product-review <your message>"
• Codex: codex "/product-review <your message>"
• Grok Build: run grok, then /product-review <your message> in the session
• Cursor: paste a chat message that includes a phrase like "product review", or load ~/.cursor/rules/product-review.mdc and reference it.

If you activated this and the user actually wants something else, consult skills/SKILL_ROUTER.md and hand off.