Install
$ agentstack add skill-josearron-skills-ux-review ✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
About
UX Review
Walk a built UI against known heuristics and report the violations — ranked, located, each with a fix. You are evaluating, not redesigning: the output is a list of problems worth fixing, not a new design. Hand the fixes to /ux-flows and /make-it-obvious.
Set up the evaluation
- Know who you're evaluating for. Pull the persona and mental model from
USER-BRIEF.md(/understand-the-user) if it exists; otherwise state the assumed user in one line. A heuristic violation only matters relative to a user trying to do something. - Walk the real thing. Run the UI if you can and go through the flow state by state — empty, loading, success, error, edge. If you can't run it, review the markup and components. Never review from memory of how it "should" look.
The heuristics
Check every screen and transition against each. Most are violated by omission — the missing feedback, the absent escape hatch.
- Visible status — the system always shows what's happening; every action gets feedback where it happened.
- Match the user's model — words, icons, and flow match the user's world, not the database schema or the org chart.
- Obvious & scannable — actionable things look actionable; the screen is self-evident at a glance (
/make-it-obvious). - Control & escape — undo, back, cancel everywhere; no dead ends, no trap states.
- Consistency & conventions — the same thing looks and behaves the same throughout; platform conventions honored.
- Error prevention & forgiveness — slips blocked by constraints; irreversible actions confirmed or gated. Prevent the error rather than scold it.
- Recognition over recall — choices and context are visible; the user never has to remember something from a previous screen.
- Plain-language errors — failures explained in human terms with a way forward, never a raw code or a dead end.
- Orientation — every screen answers "where am I, and where can I go from here."
- Calm & minimal — nothing competes with the primary content; no needless words, no noise. Doesn't deplete goodwill (no asking for data you don't need, no dark patterns).
- Accessible — keyboard-reachable, screen-reader-labeled, sufficient contrast, never color alone for meaning.
Report
For every finding:
[SEVERITY] Heuristic — where it is
What's wrong (one line) → the fix (one line)
Severity:
- Blocker — users can't complete the task, or lose data/trust.
- Major — completable, but with real friction, confusion, or error risk.
- Minor — noticeable rough edge; works.
- Polish — cosmetic.
Done when: every screen state has been walked against all eleven heuristics, findings ranked by severity (blockers first), each located and paired with a fix, and the single highest-leverage fix called out at the end. Each blocker or major is a candidate issue for /to-issues or a task for the review-and-refactor skills.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: JoseArron
- Source: JoseArron/skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet — be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.