Passed 16 of 20 test cases with the skill, and 10 of 20 without it.
Without skill50%
With skill82%
Checks it passed that the model alone missed
Specifies bun run test:a11y as the test command.
Recommends @asym/ui Base UI combobox primitives.
Reuses @asym/ui Base UI primitives for dialog behavior.
and 5 more in the full report
7 fixed11 passed either way4 still failing0 broke
Pass-rate lift
+32 pp
Tokens
+11%
Agent turns
0%
Response time
-38%
Model gemini-3.6-flashMeasured 8/20/2026v118 of 22 cases passed with skill · 20 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 7 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 7
Prompt
We want to run the automated accessibility test suite locally across our project via the command line before submitting a pull request. What exact CLI command executes this automated check?
What was checked, without and with the skill
Checked
Without skill
With skill
Specifies bun run test:a11y as the test command.
Missed
Passed
case-07 · gemini-3.6-flash · 8/20/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md79 lines2 files
Accessibility review
Apply a focused accessibility pass without replacing Core's component system or
mistaking automated checks for conformance.
Authority and boundaries
Read docs/ai/rules/frontend.md and docs/ai/rules/testing.md first.
Reuse @asym/ui Base UI primitives for dialogs, menus, tabs, tooltips,
comboboxes, and other behavior-heavy widgets. Do not build ARIA replicas of
behavior that a shared primitive already implements.
Prefer native HTML semantics before adding roles or ARIA. ARIA supplements
semantics; it does not repair incorrect interaction behavior.
Use docs/ai/skills/components-build/SKILL.md for component API work and
docs/ai/skills/emil-design-engineering/touch-accessibility.md for Core's
touch and keyboard conventions.
Keep fixes scoped. Do not migrate UI libraries or refactor unrelated layout.
Workflow
Map the states. Identify the interactive controls, reading order,
keyboard path, loading/empty/error states, responsive variants, and any
content revealed only by hover or pointer input.
Check semantics and names. Verify landmarks, headings, lists, table
headers, labels, accessible names, descriptions, values, and state attributes.
Exercise interaction. Use Tab and Shift+Tab, Enter/Space, arrow keys where
the widget pattern requires them, and Escape for dismissible overlays. Check
visible focus, initial focus, focus containment, and focus restoration.
Check feedback. Associate helper and error text with fields, announce
material async status changes, and ensure color, toast, or hover is never the
only carrier of essential information.
Check perception. Inspect contrast, zoom/reflow, touch targets, alt text,
captions where applicable, and reduced-motion behavior. Preserve Core's
global reduced-motion baseline instead of redeclaring it in an app.
Verify proportionately. Run the focused automated suite, then perform the
manual checks automation cannot prove. If asked only to review, report exact
file:line evidence and concrete fixes without editing source.