Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Browser automation powers web testing, scraping, and AI agent interactions. The difference between a flaky script and a reliable system comes down to understanding selectors, waiting strategies, and anti-detection patterns. This skill covers Playwright (recommended) and Puppeteer, with patterns for testing, scraping, and agentic browser control. Key insight: Playwright won the framework war. Unless you need Puppeteer's stealth ecosystem or are Chrome-only, Playwright is the better choice in 202
.claude/skills/dokhacgiakhoa-browser-automation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | -6% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | -35% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | -7% | 0% |
You are a browser automation expert who has debugged thousands of flaky tests and built scrapers that run for years without breaking. You've seen the evolution from Selenium to Puppeteer to Playwright and understand exactly when each tool shines.
Your core insight: Most automation failures come from three sources - bad selectors, missing waits, and detection systems. You teach people to think like the browser, use the right selectors, and let Playwright's auto-wait do its job.
For scraping, yo
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 12,056 | 8,830 | -27% | 1 | 1 | 0% | 2,193 | 1,836 | -16% | 0 | 0 | — |
case-02 | pass→pass | 10,248 | 7,317 | -29% | 1 | 1 | 0% | 1,831 | 1,719 | -6% | 0 | 0 | — |
case-03 | pass→pass | 9,818 | 5,089 | -48% | 1 | 1 | 0% | 1,844 | 1,198 | -35% | 0 | 0 | — |
case-04 | pass→pass | 11,706 | 9,442 | -19% | 1 | 1 | 0% | 1,891 | 1,753 | -7% | 0 | 0 | — |
case-05 | pass→pass | 10,156 | 8,536 | -16% | 1 | 1 | 0% | 1,798 | 1,788 | -1% | 0 | 0 | — |
case-06 | pass→pass | 14,549 | 14,654 | +1% | 1 | 1 | 0% | 2,420 | 2,831 | +17% | 0 | 0 | — |
case-07 | pass→pass | 19,155 | 17,578 | -8% | 1 | 1 | 0% | 2,688 | 3,453 | +28% | 0 | 0 | — |
case-08 | pass→pass | 11,400 | 12,600 | +11% | 1 | 1 | 0% | 2,027 | 2,198 | +8% | 0 | 0 | — |
case-09 | pass→pass | 11,230 | 11,470 | +2% | 1 | 1 | 0% | 2,036 | 2,656 | +30% | 0 | 0 | — |
case-10 | pass→pass | 13,537 | 10,808 | -20% | 1 | 1 | 0% | 2,451 | 2,434 | -1% | 0 | 0 | — |
case-11 | pass→pass | 8,814 | 5,169 | -41% | 1 | 1 | 0% | 1,656 | 1,232 | -26% | 0 | 0 | — |
case-12 | pass→pass | 12,271 | 12,335 | +1% | 1 | 1 | 0% | 2,148 | 2,615 | +22% | 0 | 0 | — |
case-13 | pass→pass | 10,136 | 8,509 | -16% | 1 | 1 | 0% | 1,819 | 1,843 | +1% | 0 | 0 | — |
case-14 | pass→pass | 15,753 | 14,990 | -5% | 1 | 1 | 0% | 2,769 | 3,185 | +15% | 0 | 0 | — |
case-15 | pass→pass | 9,377 | 4,419 | -53% | 1 | 1 | 0% | 1,567 | 1,070 | -32% | 0 | 0 | — |
case-16 | pass→pass | 11,281 | 7,000 | -38% | 1 | 1 | 0% | 1,939 | 1,604 | -17% | 0 | 0 | — |
case-22 | pass→pass | 4,151 | 4,064 | -2% | 1 | 1 | 0% | 720 | 1,024 | +42% | 0 | 0 | — |
case-17 | pass→pass | 8,327 | 5,421 | -35% | 1 | 1 | 0% | 1,314 | 1,204 | -8% | 0 | 0 | — |
case-18 | fail→pass | 19,557 | 21,435 | +10% | 1 | 1 | 0% | 2,995 | 3,729 | +25% | 0 | 0 | — |
case-19 | pass→pass | 10,760 | 9,947 | -8% | 1 | 1 | 0% | 1,826 | 1,890 | +4% | 0 | 0 | — |
case-20 | pass→pass | 14,886 | 15,309 | +3% | 1 | 1 | 0% | 2,877 | 3,122 | +9% | 0 | 0 | — |
case-21 | pass→pass | 19,335 | 16,991 | -12% | 1 | 1 | 0% | 3,465 | 3,213 | -7% | 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 +9 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.