콘텐츠로 이동

Website Console real pilot

Source

This internal pilot used the Design AI Website Console in the current repository. The project owner authorized evidence collection, the approved file selectors, the responsive navigation change, browser verification, commit, and push. The P6-P12 chain is preserved under evidence/p12/; the original review workflow and scope approval remain independently hashable.

Change

On a 390-pixel viewport, nine section buttons formed a tall navigation block that pushed the active work surface down the page. The approved change keeps those buttons in one native horizontal scroll row, preserves a 44-pixel target height, and returns keyboard focus to the active button after each render. Desktop keeps its existing sidebar.

The focus outline is a solid three-pixel #4f46e5 line. Its measured contrast is 6.29:1 on white and 5.62:1 on the active navigation surface, above the 3:1 non-text contrast requirement.

Pilot measures

Measure Recorded result
Time to first useful P6 artifact 14 seconds
Findings accepted / rejected / unresolved 1 / 0 / 0
Approval gates 10 total; 3 approved, 4 not required, 3 pending release gates
Implementation evidence Complete
Unresolved implementation risk 0

The three pending gates are commit, push, and the declared external write. They are release controls, not missing implementation evidence.

Verification

npm run release:check passed with 815 tests, all 8 strict audits, 0 documentation warnings and errors, 766 packaged files, and installed-bin plus one-shot npm exec package smoke. The focused Website Console suite passed 25 tests.

Playwright verified 1440x1000 and 390x844 viewports. Mobile body width remained 390 pixels with no page overflow; navigation stayed 52 pixels tall in one row; every target was 44 pixels tall. Forward and reverse keyboard navigation retained focus after activation, five repeated activations stayed stable, reduced-motion preference was detected, and the console reported no errors or warnings. A manual screen-reader session was not run.

Website Console imported a 416,114-byte P12 artifact and exported the exact original bytes. Input, browser storage, and export all share SHA-256 0491b4f2ea369b102d3faffb4dffab0dad91c8c8e046120910ef663694af597a. Clearing P12 restored the exact P11 implementation-evidence stage.

Permission boundary

The implementation changed only the approved Website Console source, focused test, case, status documents, and evidence/p12/**. No dependency, migration, deployment, or production system changed. Browser verification used a loopback preview. Commit and push remain separate recorded release actions.

Remaining risk

The final nested P12 MCP response is 452,923 bytes, so the default full response still returns an explicit OUTPUT_TOO_LARGE error at the 220,000-byte boundary. A fresh stdio process verified that compact: true returns 4,021 bytes with all three source hashes and byte counts unchanged, plus the same measures, issues, claims, next action, and safety boundaries. Integrators must request compact mode for large chains and retrieve the referenced full artifacts separately when they need nested source bodies.

Claim boundary

Class What this case supports
Real One consented internal review-to-implementation run and its exact local evidence
Synthetic Packaged fixture and smoke checks, kept separate from the pilot
Inferred The workflow is operable by this repository's owner under the recorded controls
Unverified External identity, authentic customer feedback, adoption, production quality, business impact, and manual screen-reader behavior

This case proves one internal workflow. It is not a customer-adoption or production- outcome claim.