About this project

Web UI Quality (WUQ) is an open-source preview of a verification layer for AI-assisted Web changes. Its stated purpose is to help a person check an AI-generated Web change before calling it finished: what the AI was allowed to change, what actually changed, what was verified, and what remains unknown. How it works The tool keeps four questions apart: Scope, Change, Evidence and Verify. Scope covers which project and files may be considered; Change covers what a Host actually applied; Evidence covers what tests, tools, a browser and before/after records really ran; Verify covers whether that evidence is sufficient for the requested claim. A repair requiring a source change must receive the applicable Host write receipt and current verification evidence before it can be reported as VERIFIED. The project states that it does not silently install dependencies, upload source, commit, push, deploy, or convert a model suggestion into write authority. Installation and commands The current preview is not published to PyPI or npm, so it is installed from a source checkout: clone the repository, run pip install -e ., then invoke web-ui-quality doctor. Stable entry points include web-ui-quality with --help, run, doctor and auth subcommands. A run takes a local Web project path plus a task described in plain language; the README's example asks to check the mobile layout of a checkout page while forbidding changes to login logic. For integrations that should fail when a requested repair claim is not verified, --ci --require VERIFIED is available, and does not bypass Host approval, scope checks, browser requirements or project-tool evidence. Self-contained public checks are run from a checkout with python -B run_tests.py. Demonstration fixture The repository includes examples/scope-drift-demo/, a small public fixture for a bounded component repair. The request names one component file while a companion file is deliberately outside that scope. In a normal source-only checkout the command records a real before state and stops until a trusted Host applies the change and returns the required evidence; if a Host modifies the companion file outside the approved scope, the project-drift gate prevents the result from becoming VERIFIED. The README describes this as a boundary demonstration rather than a claim about model accuracy or production behaviour. Status and stated limits The current public release is identified as v4.3.0-oss-preview.1 under Apache-2.0; a historical v4.3.0 release is described as a separate MIT-licensed source snapshot retained as history. The status table marks public source and engineering checks as PASS, licensing for the current preview as Apache-2.0, PyPI/npm as not published, real Host qualification as NOT_MEASURED, external blind holdout as NOT_MEASURED, and commercial GA qualification as NOT_ELIGIBLE. The preview explicitly does not claim real accuracy for Codex, Claude or other Hosts, real-user benefit, production compatibility, universal accessibility, remote attestation or token savings, and missing browser or Host evidence stays unverified. Repository layout and licensing boundary The runtime and CLI live under runtime/python/web_ui_quality/, with skills/ for Codex skill instructions, schemas/ for public JSON schemas, examples/ for local fixtures and qualification inputs, and tests/public/ for tests intended to run from a clean checkout. Publication boundary documents include PUBLICATION.md, DEPENDENCY_LICENSE_REVIEW.md and FINAL_PUBLIC_SOURCE_MANIFEST.json. Private recovery records, chat history, browser storage, evolution/, external worktrees, commercial templates and commercial demo material are excluded from this public core. Security reports are requested through a GitHub security advisory or a project contact address rather than a public issue, and the licensing boundary is described in LICENSE, NOTICE.md and THIRD_PARTY_NOTICES.md.