Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Browser automation via Chrome/Chromium CDP — open, snapshot, click, screenshot. For testing web apps, mobile layouts, and automated interactions without Playwright/Puppeteer.
.claude/skills/x-cmd-agent-browser/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -69% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 123% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -26% | 0% |
x env use agent-browser
npm install -g agent-browser
brew install agent-browser
agent-browser install # optional, skip if using existing Chrome"/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" --remote-debugging-port=9222
agent-browser --cdp 9222 open <url>Use --profile /tmp/ab-default to authorize once and share across sessions. Each new --user-data-dir triggers a Chrome auth prompt — reuse the default to avoid it.
bashagent-browser --profile /tmp/ab-default open <url> --session myapp-desktop --headed agent-browser open <url> --session myapp-iphone14 agent-browser --session myapp-iphone14 set device "iPhone 14"
Session names are global — use <project>-<device> to avoid collisions across agents. Sessions share the daemon but are fully isolated (cookies, cache, tabs).
agent-browser skills get core --full
agent-browser open <url> --session <name> --device "iPhone 14"
agent-browser snapshot -i
agent-browser click @e<id>
agent-browser screenshot
agent-browser close --allData accumulates — see CLEANUP.md for periodic cleanup.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 15,323 | 2,532 | -83% | 1 | 1 | 0% | 2,386 | 745 | -69% | 0 | 0 | — |
case-02 | fail→pass | 12,765 | 7,420 | -42% | 1 | 1 | 0% | 2,302 | 1,515 | -34% | 0 | 0 | — |
case-03 | fail→pass | 5,463 | 6,644 | +22% | 1 | 1 | 0% | 585 | 1,303 | +123% | 0 | 0 | — |
case-04 | pass→pass | 13,406 | 4,687 | -65% | 1 | 1 | 0% | 1,931 | 1,111 | -42% | 0 | 0 | — |
case-05 | pass→pass | 10,877 | 8,950 | -18% | 1 | 1 | 0% | 1,460 | 904 | -38% | 0 | 0 | — |
case-06 | pass→pass | 9,296 | 4,286 | -54% | 1 | 1 | 0% | 1,366 | 900 | -34% | 0 | 0 | — |
case-07 | fail→pass | 11,618 | 8,594 | -26% | 1 | 1 | 0% | 1,809 | 1,255 | -31% | 0 | 0 | — |
case-08 | fail→pass | 16,816 | 10,363 | -38% | 1 | 1 | 0% | 2,522 | 1,862 | -26% | 0 | 0 | — |
case-09 | pass→pass | 13,093 | 4,821 | -63% | 1 | 1 | 0% | 2,007 | 1,040 | -48% | 0 | 0 | — |
case-10 | pass→pass | 9,188 | 3,165 | -66% | 1 | 1 | 0% | 1,443 | 795 | -45% | 0 | 0 | — |
case-11 | fail→pass | 27,089 | 4,350 | -84% | 1 | 1 | 0% | 2,755 | 1,067 | -61% | 0 | 0 | — |
case-12 | fail→pass | 9,267 | 2,092 | -77% | 1 | 1 | 0% | 1,482 | 636 | -57% | 0 | 0 | — |
case-13 | pass→pass | 9,392 | 6,176 | -34% | 1 | 1 | 0% | 1,435 | 790 | -45% | 0 | 0 | — |
case-14 | pass→pass | 8,417 | 6,804 | -19% | 1 | 1 | 0% | 1,389 | 736 | -47% | 0 | 0 | — |
case-15 | fail→pass | 28,818 | 3,160 | -89% | 1 | 1 | 0% | 2,010 | 831 | -59% | 0 | 0 | — |
case-16 | fail→pass | 11,144 | 2,669 | -76% | 1 | 1 | 0% | 1,448 | 678 | -53% | 0 | 0 | — |
case-17 | pass→pass | 10,249 | 4,542 | -56% | 1 | 1 | 0% | 1,862 | 1,141 | -39% | 0 | 0 | — |
case-18 | fail→pass | 12,625 | 2,594 | -79% | 1 | 1 | 0% | 1,822 | 733 | -60% | 0 | 0 | — |
case-19 | fail→pass | 15,418 | 2,283 | -85% | 1 | 1 | 0% | 2,541 | 647 | -75% | 0 | 0 | — |
case-20 | fail→pass | 30,049 | 2,550 | -92% | 1 | 1 | 0% | 2,619 | 623 | -76% | 0 | 0 | — |
case-21 | pass→pass | 7,142 | 21,977 | +208% | 1 | 1 | 0% | 1,099 | 1,556 | +42% | 0 | 0 | — |
case-22 | pass→pass | 11,148 | 21,171 | +90% | 1 | 1 | 0% | 1,872 | 1,968 | +5% | 0 | 0 | — |
case-23 | pass→pass | 8,811 | 5,848 | -34% | 1 | 1 | 0% | 1,683 | 1,356 | -19% | 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. 23 cases were attempted. The headline lift of +52 percentage points is the difference between those two pass rates over the 23 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.