---
name: begin-quality
description: Review and repair a website's actual rendered design, content, interactions, responsive behavior, and core action using proportionate checks and clear observations.
---

# Begin quality

Review the website against its brief and the requested change. Read [review and repair](references/review.md) for choosing checks, inspecting rendered behavior, and reporting useful results.

Discover the project's real commands and available browser tools. Inspect the preview at the smallest supported width, around layout breakpoints, and at a representative desktop size when possible. Measure horizontal overflow and comfortable touch targets for navigation and main actions. Exercise those actions and important error/recovery states. Select checks that establish the behavior being changed rather than running every possible suite; unavailable browser access leaves rendered checks unverified.

Review design and English content together: reading order, hierarchy, spacing, real text, imagery, useful actions, and the owner's intended voice. Identify concrete defects and explain aesthetic suggestions as judgments. Do not equate a passing build with a visually reviewed site.

Separate a poor typeface choice from poor execution of a suitable choice. Review populated results as carefully as their opening pages: full names, quantities, rules, shared fact baselines and meaningful headings. Check that page reading reaches all result content; nested scrolling can hide material without page overflow. Inspect assembled titles for unnecessary terminal full stops. Inspect supporting lines for repeated meaning that delays the actual content; retain useful conditions and instructions. Check custody, dates, chosen routes and recovery actions at their actual rendered size; a quiet utility role still needs comfortable reading.

For a sequence, dependency or measurement view, compare the visible relationship with the actual records. Inspect consequential names, timing and units at their rendered size, connector destinations in populated states, and normal phrase spacing when desktop breaks disappear. A correct calculation or overflow measurement does not establish those reading qualities.

Read assembled guide headings and generated results at more than one narrow width when their wrapping differs. Trace unexpected gaps through actual markup and inline styles, then inspect the repaired phrase and its line spacing. At the end of a long reading, check that its promised next action remains findable. Review raw-text units and real keyboard focus separately from the main structured result.

Candidate font previews leave the complete composition unverified. Match a dominant diagram's connections and grouping to the actual records and the visitor's task. Inspect the assembled text frame and transformed art; check minimum-height/aspect-ratio controls around their actual breakpoints. On a phone, inspect how the visitor moves between editing, current results, prepared output and local recovery. Open actual downloaded artifacts when their rendering, physical dimensions or embedded fonts are part of the promise.

For a design critique, a disappointing result, or a request for high visual quality, read [visual review](references/visual-review.md). Inspect the whole composition and the pages visitors actually use. A useful critique names what weakens the result and its effect; it can justify regrouping or reordering content rather than only adjusting CSS values. Use a fresh review perspective when available and useful, without making separate Agents a requirement.

For a defect, preserve the current work, trace its cause, apply a focused repair, and rerun the relevant check. After checks pass, broaden testing only for an unresolved issue or a new change. Record passed, failed, not-run, and unknown observations separately; capture user satisfaction or adoption only when the person actually states it.