AgentSkillsCN

audit

对代码库的某一部分执行结构化的产品/UX 审计。适用于评估 UI 质量、检查回归问题,或定期开展产品评审时使用。当用户说出“审计”、“审视 UX”、“检查问题”或“产品评审”等关键词时,自动触发该技能。

SKILL.md
--- frontmatter
name: audit
description: "Run a structured product/UX audit on a section of the codebase. Use when reviewing UI quality, checking for regressions, or doing periodic product reviews. Auto-invoke when the user says 'audit', 'review the UX', 'check for issues', or 'product review'."

Product & UX Audit Skill

Step 1: Scope

Ask the user which area to audit, or audit the area they've specified. Read all relevant component files, styles, and related hooks.

Step 2: Systematic Review

Check each component/page for:

Visual Consistency

  • Light mode AND dark mode rendering (check for white slabs, missing transparency, contrast issues)
  • Consistent spacing, typography, and color usage across components
  • Ambient vs. opaque backgrounds — flag any sterile/boxy layouts

Functional Completeness

  • Empty states handled
  • Loading states present
  • Error states handled gracefully
  • Responsive behavior

Data Integration

  • Hooks returning expected data
  • RLS policies not hiding intended content
  • Edge cases with missing or null data

UX Quality

  • Interactive elements have proper hover/focus states
  • Navigation is intuitive
  • Actions are discoverable (not hidden)

Step 3: Report

Save findings to docs/audits/<area>-audit-<date>.md with:

P0 — Critical (fix immediately)

[Issues that break functionality or severely degrade UX]

P1 — Important (fix this sprint)

[Issues that noticeably hurt quality]

P2 — Nice to Have (backlog)

[Polish items and minor improvements]

Step 4: Ask

Ask the user if they want to implement any P0 fixes immediately.