accessibility-coverage-matrix

Accessibility Coverage Matrix

Inputs

• scope: Required. Repository or target scope to assess.
• frameworks: Optional. Comma-separated framework set to evaluate. Defaults to wcag-22,aria-apg,coga,section-508,en-301-549.
• mode: Optional. Matrix execution mode. Allowed values are build, refresh, report, or probe. Defaults to build.
• baseUrl: Optional. Base URL for runtime probing. If omitted, derive it from the target configuration or ask the user.
• serve: Optional. Serve mode for the harness. Allowed values are auto or external. Defaults to auto.

Core model

Treat the matrix as a criterion x surface x method grid.

• Criterion: framework-specific success criteria or control identifiers.
• Surface: discrete UI surfaces such as a page, component, widget, global chrome, or content type.
• Method: evidence method such as static-source, axe-auto, runtime-automation, manual-keyboard, cognitive-walkthrough, screen-reader, or other method names recorded by the engine.
• Cell lifecycle: not-started -> blocked, partial, fail, pass, or not-applicable based on newly ingested evidence and method adequacy.
• Method-adequacy semantics: a cell is counted as covered only when the winning evidence method is one of the criterion's adequateMethods in the reviewed criteria catalog. A pass from an inadequate method does not count as covered. The probe-criteria-map records which criteria a probe reports on. It never authorizes coverage on its own.

Required steps

1. Bootstrap or resume the matrix under .copilot-tracking/accessibility/coverage/ and create or update the working artifacts for the target scope.
2. Load the criteria catalog and adequateMethods from the accessibility skill framework references before evaluating any cells.
3. Delegate to the Accessibility Surface Inventory subagent as the sole producer of a11y-runtime.config.json. Do not author that config yourself. Pause for the user to review or override it before proceeding.
4. Build the grid with the runtime_a11y matrix engine, using the loaded framework and surface definitions.
5. Ingest existing and static evidence as data, including assessor findings, planner state.json data, prior reports, and prior matrix artifacts. Preserve provenance and do not invent evidence.
6. Load the accessibility skill and run the runtime probe harness from that skill's root, which holds the uv project, using its script entrypoint:
   • uv run scripts/runtime_a11y/__main__.py run-all --config a11y-runtime.config.json --out results.json for normal runs.
   • uv run scripts/runtime_a11y/__main__.py probe <probeId> --config ... for probe mode.
   • From another working directory, pin the project instead: uv run --project <skill-root> <skill-root>/scripts/runtime_a11y/__main__.py ....
   • Install the skill-local Node dependencies once with npm ci in scripts/runtime_a11y/ under the loaded skill root before the first run.
   • Add --trace when a trace is needed and --allow-external only when the target is an approved non-loopback host after explicit confirmation.
7. Route fail, partial, or blocked cells through the Finding Deep Verifier. Involve the Codebase Profiler and Accessibility Framework Assessor by role as needed to interpret findings and close gaps.
8. Compute coverage, residual gaps, and nextActions with the engine, then write the matrix JSON and render the canonical evidence bundle. When a cell still requires human-led assistive-technology evidence, route the tester to the shared real screen reader testing runbook instead of writing case-specific manual instructions inline.
   • uv run scripts/runtime_a11y/__main__.py render-artifacts --matrix coverage-matrix-{repo-slug}.json --output-dir .copilot-tracking/accessibility/coverage --repo-slug {repo-slug}
   • Preserve every file listed by the generated artifact manifest. Do not hand-author the EARL or manual test-plan files.
9. Present the coverage summary, EARL result counts, pending manual test count, and artifact manifest path. The Markdown artifacts retain the canonical accessibility disclaimer and unchecked human-review checkbox.