catch-ux-risks

Catch UX Risks

Use this skill to review a product surface before launch.

The output is a product-risk artifact, not validated user research. Do not invent user quotes, survey findings, or behavioral certainty.

Inputs To Request Or Use

• Product surface: URL, screenshot, PR, prototype, spec, or flow description
• Target audience or primary user job
• Launch context or product stage
• Known constraints, if supplied

If the surface is thin, state the evidence limitation and keep confidence lower.

Repo-Backed Mode

If the current workspace is a Preflight UX checkout, or PREFLIGHT_UX_REPO points to one, prefer the repository artifacts over freeform review:

• Confirm the checkout with bundled script python3 scripts/preflight_ux.py locate.
• Validate repo structure with python3 scripts/preflight_ux.py validate when the task depends on repo artifacts.
• Use panels/default.yaml or a supplied panel file for review lenses.
• Use taxonomy/issue-classes.yaml for issue-class names.
• Use schemas/report-finding.schema.json for normalized finding shape.
• If the surface is already saved as Markdown, scaffold a run with python3 scripts/preflight_ux.py uxpanel run --surface <surface.md> --surface-type spec --panel panels/default.yaml --run-id <slug>.
• If the user wants a benchmark comparison rather than a product review, switch to the calibrate-ux-findings skill after findings are produced.

If no checkout is available, still perform the review using the structure below and mention that repo validation was not run.

Review Lenses

Look for concrete risks in these classes:

• comprehension: users cannot tell what to do, what changed, or why it matters
• trust: output, source, policy, data, or provenance is not credible enough
• workflow: the main task is blocked, slowed, hidden, or made harder
• recovery: errors, empty states, loading, fallback, or support paths are weak
• inclusion: users with higher support needs or fewer options are disadvantaged
• operator fit: density, shortcuts, navigation, or repeated-use flows are mismatched
• change regression: a redesign removes expected behavior or familiar structure

Output Format

Return 5-10 findings as a table:

Priority	Issue class	Risk hypothesis	Surface evidence	Recommendation	Confidence

Then add:

• Likely misses: what this review may not catch
• Validation plan: how to test the highest-risk findings with real evidence
• False-positive watch: which findings are most likely to be over-predictions

Rules

• Tie every finding to supplied surface evidence.
• Prefer specific task blockers over generic UX advice.
• Separate severity from confidence.
• Mark unsupported but plausible concerns as low confidence or omit them.
• Never claim the output proves real user behavior.