Installs into .claude/skills of the current project.
Are you the author of Ux Heuristic Evaluation?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/eliyce-ux-heuristic-evaluation)
---
name: ux-heuristic-evaluation
description: Evaluate UI behavior against usability heuristics.
model_tier: medium
triggers:
- ui_impact:
- new-component
- new-screen
- redesign
cacheable: true
cache_key_inputs:
- docs/modules/*/ui/components.md
- docs/modules/*/ui/screens.md
output_format: markdown
input_schema:
ui_surface_paths:
type: path[]
required: true
description: UI surfaces or docs to evaluate.
request_text:
type: string
required: false
description: Optional evaluation scope.
---
## What It Does
Evaluates changed UI behavior against a defined heuristic rubric so usability issues are reported as concrete user-impact findings instead of subjective design opinions.
## Use This When
Use this for any meaningful UI change once the target flow and states are known, especially before final review or documentation handoff.
## Inputs
- Read the affected UI docs, flow notes, and current state inventory first.
- Read `references/heuristic-rubric.md` before evaluating the interface.
- Read the relevant stack-specific browser validation guide if behavior needs interactive confirmation.
## Procedure
1. Walk every heuristic in `assets/heuristics.txt` against the changed interface; never skip a category silently.
2. Tie each issue to a user task, state, or observed confusion point — never style preference.
3. Bucket findings into Blocking Issues vs Improvement Opportunities per `assets/output.template.md`.
4. Note any UI doc that must update because the heuristic reveals missing state documentation.
5. Validate with `scripts/lint-output.sh`.
## Output Contract
- Return sections named `Blocking Issues` and `Improvement Opportunities`.
- Each finding must state the affected screen or component, the heuristic violated, and the user impact.
- If no issues are found, write `Blocking Issues: none` and `Improvement Opportunities: none`.
## Escalate / Stop Conditions
- Ask when the expected user goal or task completion path is unclear.
- Warn when the interface appears to conflict with accessibility or recovery expectations.
- Do not file purely stylistic preferences as heuristic issues.
## Resources
- `references/heuristic-rubric.md`
- `scripts/lint-output.sh`
- `assets/output.template.md`
- `assets/heuristics.txt`
- `runtime/capabilities/coding/stacks/flutter/references/tools/playwright.md`
- `agents/openai.yaml`