| name | agent-ux-audit |
| description | (Codex) Strategic UX audit using discovery gate, task flow mapping, and targeted evaluation. Focus on impact over completeness. |
UX Audit
Use this skill for a strategic UX review of the app. Focus on finding the right high-impact issues, not all issues.
For full audits, use this as the authoritative workflow. Stop redundant auditing once a pattern is confirmed across 2-3 instances.
Tool Access
Browser automation requires Playwright (playwright-cli). If Playwright is not available, stop and tell the user to install it before proceeding.
If you need any tool that is not available (e.g. gh CLI not installed, Playwright not configured, a service unreachable), stop and tell the user what tool you need, why, and what they should do to provide access. Do not invent ad-hoc alternatives.
Core Principles
- Your goal is to find the right issues, not all issues.
- Skip checklist items that don't apply to a page's primary task.
- Stop redundant auditing once a pattern is confirmed across 2-3 instances.
- Depth-dive into critical blockers; keep polish findings brief.
- Connect findings to user impact: who, how often, what happens.
- Accessibility is usability. Violations are UX failures, not compliance theater.
Core Rules
- Do not modify source files during the audit.
-
-
- Use the actual deployed URL provided by the user.
- Do not start a local server and do not switch to localhost during a full audit.
- Do not audit pages in isolation — map end-to-end workflows and cross-page task friction.
- Check both desktop and mobile when mobile is in scope.
- Capture snapshots or screenshots before evaluating pages.
- Validate authenticated state before treating protected routes as valid.
- Deduplicate issues by pattern, not by isolated page instance.
- If a route requires specific data state, file the issue and note the missing required state instead of skipping.
Full Audit Flow
- Authenticate for the target environment:
- Read the router source files to discover all routes.
- Select major routes from the discovered list.
- For each major route, spawn a focused sub-task to:
- Navigate to the page using Playwright on the deployed URL.
- Capture screenshots.
- Evaluate layout, spacing, responsiveness, and accessibility against the design system and UX skills.
- File issues using the project's issue format with appropriate labels.
- Note required data state if the page cannot be fully exercised due to missing records.
- After all routes finish, write a summary file grouping issue links by severity.
-
Required Companion Skills
Synchronization Contract
- Canonical deep UX criteria live in
.claude/skills/ux-audit/*.md.
.codex/skills/ux-audit-* files are concise operational summaries and must remain semantically aligned with canonical files.
Output
- Findings grouped by severity and pattern
- Evidence for each finding
- Issue links created with required labels
- A summary file grouped by severity
