← ClaudeAtlas

web-experience-visual-qualitylisted

Evaluate browser-observed web experience quality across critical states, responsive cases, accessibility observations and visual discrepancy iterations. Use for material user-visible web UI changes and preview/release quality claims.
evoliainfo/senior-engineering-for-ai-agents · ★ 1 · Web & Frontend · score 64
Install: claude install-skill evoliainfo/senior-engineering-for-ai-agents
# Web Experience & Visual Quality Use this pack when the mission has a material user-visible web surface. Do not load it for purely backend or non-visual changes. ## What this pack does It turns real browser/visual observations into a deterministic quality decision. It does not pretend that a code review or model opinion proves rendered UI quality. ## Before evaluation Declare only the cases that matter to the current product slice. Each case identifies: - the critical UI state or journey checkpoint; - the viewport class to exercise; - whether accessibility evidence is required. Prefer a small set of meaningful cases over an exhaustive screenshot catalog. ## Observe with real tools Resolve `browser` and `visual_capture` through the active harness. M4 owns tool resolution; this pack never invents availability or credentials. For each required case, collect: - browser interaction status; - screenshot or equivalent rendered visual reference; - a `capture_context_ref` describing the concrete browser/rendering conditions used for that evidence; - accessibility observation/reference when required; - any discovered visual discrepancy with stable id, severity and resolution state. Visual screenshot baselines must be compared under sufficiently stable rendering conditions. A bare `capture_stable=true` is not enough: the observation also needs capture-context evidence. If the environment is not comparable, mark the case incomplete rather than accepting a noisy result. ## E