Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Interactive screen reader simulation for education and debugging. Parses HTML/JSX and produces step-by-step narration of what a screen reader would announce. Supports reading order, Tab navigation, heading navigation, and form navigation modes.
.claude/skills/community-access-screen-reader-lab-733cd5/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | -7% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | 40% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | 43% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 70% | 0% |
You are a screen reader simulation agent. You parse HTML/JSX and produce a step-by-step narration of what a screen reader would announce, helping developers understand the accessible experience.
Disclaimer: This is an educational simulation. Real screen reader behavior varies between NVDA, JAWS, VoiceOver, and Narrator.
Walk the DOM in reading order. For each element announce: role, accessible name, state, description.
Simulate Tab through focusable elements. Flag focus traps, unreachable interactive elements.
List all headings by level. Flag skipped levels, missing H1, multiple H1s.
List form controls with labels. Flag unlabeled inputs, missing required indicators.
aria-labelledby → 2. aria-label → 3. Native <label> → 4. Element content → 5. title → 6. placeholder| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 6,498 | 4,805 | -26% | 1 | 1 | 0% | 1,177 | 1,100 | -7% | 0 | 0 | — |
case-02 | pass→pass | 4,119 | 4,748 | +15% | 1 | 1 | 0% | 771 | 1,077 | +40% | 0 | 0 | — |
case-03 | pass→pass | 5,023 | 5,446 | +8% | 1 | 1 | 0% | 824 | 1,176 | +43% | 0 | 0 | — |
case-04 | pass→pass | 4,112 | 4,884 | +19% | 1 | 1 | 0% | 603 | 1,023 | +70% | 0 | 0 | — |
case-05 | pass→pass | 9,957 | 16,501 | +66% | 1 | 1 | 0% | 1,785 | 3,289 | +84% | 0 | 0 | — |
case-06 | fail→pass | 7,314 | 6,954 | -5% | 1 | 1 | 0% | 1,243 | 1,458 | +17% | 0 | 0 | — |
case-07 | pass→pass | 3,998 | 3,298 | -18% | 1 | 1 | 0% | 671 | 803 | +20% | 0 | 0 | — |
case-08 | pass→pass | 9,759 | 4,785 | -51% | 1 | 1 | 0% | 1,669 | 1,092 | -35% | 0 | 0 | — |
case-09 | pass→pass | 3,952 | 3,363 | -15% | 1 | 1 | 0% | 655 | 792 | +21% | 0 | 0 | — |
case-10 | pass→pass | 11,341 | 6,158 | -46% | 1 | 1 | 0% | 1,847 | 1,398 | -24% | 0 | 0 | — |
case-11 | pass→pass | 13,266 | 10,090 | -24% | 1 | 1 | 0% | 2,149 | 2,047 | -5% | 0 | 0 | — |
case-12 | pass→pass | 9,356 | 6,850 | -27% | 1 | 1 | 0% | 1,637 | 1,597 | -2% | 0 | 0 | — |
case-13 | pass→pass | 13,078 | 7,096 | -46% | 1 | 1 | 0% | 2,180 | 1,420 | -35% | 0 | 0 | — |
case-14 | pass→pass | 7,107 | 4,753 | -33% | 1 | 1 | 0% | 1,275 | 1,166 | -9% | 0 | 0 | — |
case-15 | pass→pass | 9,473 | 5,454 | -42% | 1 | 1 | 0% | 1,610 | 1,243 | -23% | 0 | 0 | — |
case-16 | pass→pass | 4,901 | 3,423 | -30% | 1 | 1 | 0% | 944 | 869 | -8% | 0 | 0 | — |
case-17 | pass→pass | 8,371 | 3,776 | -55% | 1 | 1 | 0% | 1,655 | 1,044 | -37% | 0 | 0 | — |
case-18 | pass→pass | 5,089 | 5,710 | +12% | 1 | 1 | 0% | 909 | 1,242 | +37% | 0 | 0 | — |
case-19 | pass→pass | 10,753 | 12,896 | +20% | 1 | 1 | 0% | 1,893 | 2,675 | +41% | 0 | 0 | — |
case-20 | pass→pass | 12,512 | 10,091 | -19% | 1 | 1 | 0% | 2,498 | 2,455 | -2% | 0 | 0 | — |
case-21 | pass→pass | 21,445 | 20,011 | -7% | 1 | 1 | 0% | 3,439 | 3,515 | +2% | 0 | 0 | — |
case-22 | pass→pass | 14,284 | 10,308 | -28% | 1 | 1 | 0% | 2,895 | 2,372 | -18% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +5 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.