usability-heuristic-evaluation
Guides judging a running interface as a first-time user would, without its specifications: citing a named principle for every finding, walking each task step, reading labels and addresses for scent, and rating severity.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 2
- Forks
- 0
01 前言 Front Matter
- name
- usability-heuristic-evaluation
- description
- Guides judging a running interface as a first-time user would, without its specifications: citing a named principle for every finding, walking each task step, reading labels and addresses for scent, and rating severity.
- when_to_use
- Use when running the spec-blind usability lens of an exploratory and usability review, or when deciding whether a usability observation is a finding.
- compatibility
- Requires a running interface in a real browser, and no access to its specifications, source, or designs.
03 SKILL.md 正文
Usability Heuristic Evaluation
Exploratory and Usability Review owns the sequence: the frozen tasks, the two lenses in order, structural blindness, passive conduct, and one bounded pass. Usability Probes and Completeness owns the probes every pass runs, and Accessibility owns the full accessibility audit. This skill covers the evaluator's judgement inside the usability lens.
What Counts as Ground Truth
Judge only what a first-time user can perceive: rendered text, labels, layout, affordances, feedback, the address, and behaviour seen by interacting. Measure it against published usability principles, the interface's own consistency, and prevailing conventions, which an evaluator may look up. The product's intended behaviour is not ground truth: once known, first-time comprehension can no longer be judged.
No Principle, No Finding
"Confusing" is a reaction, not a finding. Every finding names what it violates: one of Nielsen's ten usability heuristics, a failed walkthrough question, a recognised interaction law, an ISO 9241-110 interaction principle, or a WCAG 2.2 Understandable criterion. When no principle is violated, record the observation as a note and move on.
Breadth, Then Depth
Sweep the whole interface against the heuristic set for breadth. Then, for depth, walk each frozen task step by step and ask at every step:
- Will the user try to achieve the right result here?
- Will the user notice that the correct action is available?
- Will the user connect that action with the result they want?
- After acting, will the user see that progress was made?
A "no" or an "unsure" is a finding. Keep the step transcript, so the verdict can be checked.
Read Like Someone Scanning
Users scan and take the first plausible option. For each key task, decide whether the page makes the right first click the most compelling target. For every link and menu item, ask whether its label alone predicts where it leads. Look past polish: attractiveness earns undeserved tolerance, and plainness alone is not a finding.
States a Demonstration Skips
Probe the empty or zero-result state, loading, errors, first visit against return visit, very long content, and slow or offline conditions. Find at least one weakness, or state that a genuine attempt found none, since silence reads as nobody trying.
Each Size Is Its Own Experience
At each viewport class, judge usability, not only layout: whether a collapsed menu can be found, whether anything disappears at one size, whether touch targets are reachable, whether reading order survives restacking, and whether one value agrees across sizes. Where sizes disagree, ask which one a newcomer would trust.
The address is interface too. A usable one reads as words, mirrors the navigation, matches its page, leaks no implementation detail, lands on a sensible parent when shortened, and follows one pattern across siblings.
Rate Severity, Not Priority
| Severity | Meaning | Criticality |
|---|---|---|
| 4 | blocks or badly misleads most users | CRITICAL |
| 3 | many users struggle; important to fix | HIGH |
| 2 | some users are slowed or briefly confused | MEDIUM |
| 1 | cosmetic; minimal impact | LOW |
| 0 | considered and dismissed | no finding |
Weigh frequency, how hard it is to overcome, and persistence across visits. Priority is the owner's decision, separate from severity.
A finding also records the task step, viewport, browser and version, locale, date, reproducibility (always, intermittent with a count, or once), reproduction steps, the expected behaviour its principle grounds, the actual behaviour with exact label text, and a suggested clarification marked as a hypothesis.
Suggesting Behaviour
When a newcomer would reasonably expect behaviour the interface lacks, propose it as a scenario only when a cited principle grounds it, it can be written as Given, When, and Then, and the product rather than a third party owns it. Mark it spec-blind: a spec-aware reviewer confirms it is not already specified. Exploratory Testing owns the spec-aware lens and its gaps.
04 同仓库更多 Skills
writing-browser-e2e-tests
Guides writing browser end-to-end tests that find elements as users do, wait on what the action produced, keep page knowledge in page objects, organize one feature per file, and run scenario-backed suites from generated specs.
usability-heuristic-evaluation
Guides judging a running interface as a first-time user would, without its specifications: citing a named principle for every finding, walking each task step, reading labels and addresses for scent, and rating severity.
tooling-nx
Guides Nx workspace work under the Nx standard: selecting projects by the graph, adding a target only where a capability exists, tracing a stale cached pass to its missing input, and answering a module-boundary finding.
programming-typescript
Guides the procedure for writing or changing TypeScript: loading the TypeScript standard first, driving each increment test-first, finding which rule a decision falls under, and confirming the gates before handing off.
programming-shell
Guides shell work under the shell standards: checking the declared dialect first, deciding whether logic belongs in shell at all, tracing where exit-on-error does not reach, and testing the script as its callers run it.
programming-javascript
Guides JavaScript work under the JavaScript standard: confirming the file is type-checked, typing through JSDoc, validating input the checker cannot see, and noticing floating promises and swallowed errors before handing off.