Surface completeness

Use this workflow during user-facing implementation, after each cosmetic change, and before declaring the affected work complete. An explicit session closeout can also cover new services and their user documentation without a UI change. Scope follows the requested change. It does not authorize unrelated redesign, releases, provider changes, or new external communication.

Preserve the accepted result

Record the accepted baseline, intended improvement, affected surfaces, and recent work that must survive in your task list or plan tool. Distinguish current source, open pull requests, approved references, and deployed screens. For overlapping pull requests, compare their incremental intent and shared history before reconciling files. Do not replace a whole file from one branch when that discards unrelated work.

A direction such as "use bolder colors" changes presentation within the established action hierarchy. It does not promote Cancel, make a safe secondary action look like success, or erase the distinction between ordinary and destructive actions. Preserve the project's accepted palette and interaction model. Use authoritative standards guidance when deciding hierarchy, action emphasis, color meaning, contrast, focus, motion, or target behavior. Separate normative accessibility requirements from design-system recommendations.

Change the shared owner

Find the existing semantic token, shared component, or style rule that owns the element. Change its parameters once and inspect all affected consumers. Avoid repeating a common style in individual components or adding a parallel system. Trace aliases, variants, generated outputs, and local overrides. A local exception needs a concrete behavior or context difference and documented scope.

When the new design changes the accepted rule, update its owning contract, tokens, and existing enforcement in the same work. Otherwise, cite the existing rule that already covers it. Do not update screenshots only to make a failure disappear. Use checks proportional to the change. Do not create persistent exact-text or CSS assertions that only mirror the implementation.

Keep one compliance and completeness reviewer

At the first cosmetic change, use one dedicated subagent or parallel worker if your client supports one, or reuse the existing reviewer. Give it the scope, accepted baseline, shared owners, affected consumer inventory, final diff, and evidence paths. Its task is independent compliance and completeness review, not redesign or autonomous repository changes. Ask it to check in at the end of each cosmetic change before that change is accepted. Do not create a new reviewer for every minor fix.

The reviewer checks:

• The result follows the accepted rules, contracts, and tokens, or the intended new rule and its enforcement were updated together.
• Shared changes propagate through their consumers. Local overrides do not defeat them, and component exceptions have a scoped reason.
• All intended surfaces, components, variants, roles, and relevant states are covered, with unrelated recent improvements preserved.
• Code documentation and user guidance reflect the finished behavior, including relevant screenshots and tutorials.

The primary worker verifies findings, repairs actual gaps, and owns the final judgment. After a repair, request focused verification of that finding instead of restarting broad review. If delegation is unavailable, record the missing gate and the checks performed. Do not describe a self-review as independent.

Prove affected coverage

Maintain a concise matrix in your task list or plan tool: shared owner, consumers, intended effect, relevant states and viewports, documentation, evidence, and unresolved gaps. Derive consumers from imports, styles, routes, registries, and existing surface inventories. Two representative screenshots alone cannot prove that a shared system change reached every consumer.

Use static consumer and override inspection to prove propagation. Use browser evidence to check distinct layouts and interactions. Include relevant phone and desktop layouts, action roles, keyboard focus, disabled, pending and error states, long content, motion, and role variants when affected. Do not demand every state for an unrelated copy or layout edit. Browser navigation uses structured snapshots and element references. Screenshots verify results and supply documentation.

Inspect intended changes before accepting visual baselines. Confirm screenshot provenance, deterministic fixtures, responsive layout, and actual user behavior. A green snapshot alone does not establish usability or complete coverage.

Choose relevant quality signals

When a change affects interaction, layout, assets, fonts, or browser features, read quality signals. Select the checks that can expose a failure in the affected user task. A copy edit does not require a full performance or device audit.

Record each relevant check's expected result, method, build or commit, browser and version, viewport or device, result, and evidence. Keep "passed", "not tested", and "not applicable" distinct. Explain exclusions. An automated accessibility scan cannot establish conformance, and a lab performance score cannot establish real-user performance.

Check browser support and a usable fallback before adopting a modern platform feature. Choose relevant keyboard and assistive-technology checks, zoom and reflow, user preferences, loading and failure states, and performance measures. Use the project's supported environments rather than a universal device list. Separate normative accessibility criteria from design and performance guidance.

Maintain documentation with the result

Map the change to code contracts, runbooks, staff guides, public design-system pages, handbooks, tutorials, screenshots, and generated architecture or status maps. Trace each displayed map to its source data. Reconcile released entries and remaining proposals with the same deployment receipt. Update relevant generated copies through their source owner. "Not affected" needs a brief reason in the matrix, not an unnecessary rewrite.

Current guides, runbooks, comments, and architecture summaries describe what the system does and how it works now. Remove obsolete behavior, superseded receipts, and development stories. Do not keep tombstone sections or historical anchors only to preserve documentation. Git history and private session evidence retain the audit trail. Keep current recovery steps and explicit verification limits. Update inbound links to support the current claim. A working anchor alone is not proof. Keep earlier test evidence separate from current deployment evidence in your task list or plan tool, and verify delivery of generated documentation.

Replace images that misrepresent the finished interface. Add screenshots when they explain a new feature, service, state, or meaningful step. Capture the final verified view with authorized content or suitable test data. Retain no secrets or unapproved private content. Preserve the useful context in approved publication assets. Label crops and illustrative recreations accurately. Check image paths, useful captions or alternative text, links, and rendered pages. Clearly label preview or undeployed behavior and keep public and staff boundaries.

Checkpoints and completion

Update your task list or plan tool after each cosmetic change, compliance review, repair, documentation update, and release boundary. Store exact source or pull request identifiers, evidence paths, findings and dispositions, and the next action so a new session can resume without repeating settled work.

At session closeout, reconcile the intended scope against the coverage matrix and final diff. Require the reviewer's final check of affected design and documentation when independent review is available. Report remaining gaps explicitly. Source, tests, merged code, deployed applications, persisted content, and public visibility are separate facts. Reuse valid evidence. This workflow never creates merge or deployment authority.

Related skills: accessibility-compliance covers accessibility implementation and testing, and web-ui-best-practices covers interface design conventions.