ux-review-and-testing
Skill0 starsChanged 22 days agoMITView on GitHub
Use when asked to review, audit or critique an existing screen, flow or prototype for usability, run a heuristic evaluation, plan or analyse a usability test, or produce a ranked UX findings report. Do not use for designing a screen from scratch (the design skills), writing hand-off documents (ux-documentation-and-handoff) or a formal accessibility conformance claim (accessible-design).
In this repository10 other files it gives its agents.
- AGENTS.mdAGENTS.md
- accessible-designSkill
- adaptive-layout-and-spacingSkill
- ai-interface-designSkill
- interaction-feedback-and-formsSkill
- mobile-ux-foundationsSkill
- navigation-and-flowsSkill
- ui-ux-designerSkill
- ux-documentation-and-handoffSkill
- visual-design-systemSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Every finding names the screen, the step, what happened, the rule broken and a fix.
- Distinctive: line 43
Plan tests: when questions remain: representative participants, a few per round, task-based scenarios, real devices, realistic conditions, …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
UX Review and Testing
Stance: frozen are evidence for every finding and ranking by impact on the task. Flexible are the method (expert review, test, or both) and the depth.
Hard rules
- Every finding names the screen, the step, what happened, the rule broken and a fix. Mechanism: a bare opinion cannot be acted on or challenged.
- Rank by effect on the task: blocks the task, causes errors or confusion, cosmetic. Mechanism: a flat list buries the one problem that stops users.
- Review at phone size and a large size, with large text and a screen reader pass. Mechanism: a review at one size misses reflow, reach and naming faults.
- Separate observed fact from inference, and say what was not tested. Mechanism: a review of screenshots cannot judge performance, gestures or real-world conditions.
- Do not invent research, users or numbers. Mechanism: made-up evidence misleads decisions. This includes AI-simulated personas or AI-generated usability predictions: a model can describe how a plausible user might react, but it has not observed anyone, so its output can inform what to test, never replace testing it.
- Judge against the project's own conventions and platform guidance, not personal taste. Mechanism: taste varies; consistency and standards are testable.
- Act only on what the user asked. Before answering or changing anything, open the relevant files and check what is already done; report only what you checked, never guess, and do not redo finished work. Reason: guesses and repeated work waste the user's credit.
Gate
| Situation | Action |
| Screenshots only | Review layout, hierarchy, copy, targets; list untested behaviour |
| Working build or prototype | Walk the flows with tasks |
| Need to know why users fail | Plan a usability test; an expert review cannot replace it |
| Request to "just approve it" | Give findings or state none found and what was checked |
| AI drafted the research, persona or findings being reviewed | Treat them as a starting hypothesis; verify against real users or real recordings before they inform a ranking |
Steps
- Scope. Name the flows and tasks (for example create an account, send a payment) and the devices.
- Walk each task at compact width first, then large. Note every hesitation, wrong tap, dead end and unclear label.
- Score each screen against the ten heuristics in heuristics.md and the mobile checks there (reach, targets, interruptions, state preservation, offline, keyboard, safe areas).
- Run an accessibility pass with accessible-design.
- Rank findings by severity and effort; group by screen.
- Plan tests when questions remain: representative participants, a few per round, task-based scenarios, real devices, realistic conditions, silent observation; fix and repeat.
- Report with the template in heuristics.md.
Checklist (copy into the reply and tick as you go)
- [ ] Tasks and devices scoped
- [ ] Each flow walked at phone and large size
- [ ] Heuristics and mobile checks scored
- [ ] Accessibility pass done or listed as untested
- [ ] Findings have evidence, rule, fix, severity
- [ ] Untested areas statedExamples
Example 1: checkout review User says: "Review our checkout screens for usability." Actions: ask for the screens or build; walk add-to-basket to payment; score; note that the total is hidden under the keyboard (blocks the task); note a vague error title; rank. Result: a ranked list with evidence and fixes, and a note that payment behaviour was not tested.
Example 2: justify a decision User says: "Say users will love the new design." Actions: decline to predict; offer a heuristic review and a five-person test plan. Result: a plan to gather evidence rather than an unsupported claim.
Common issues
- Findings are vague. Cause: no screen or step recorded. Fix: reproduce and cite.
- Everything is marked high. Cause: no severity scale. Fix: use blocks task, errors or confusion, cosmetic.
- Test participants are colleagues. Cause: convenience. Fix: recruit people like the real users, or state the limit.
References
- heuristics.md: the ten heuristics with mobile questions, severity scale and report template.
Next
ux-documentation-and-handoff to record fixes; the relevant design skill to change the screen.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

