Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Drive a real Chromium-based browser from an agent loop. Use when a task needs to navigate web pages, fill forms, scrape rendered HTML, or take screenshots.
.claude/skills/galelmalah-agent-browser/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 13 |
| gemini-3.1-pro-preview | 100% | 1 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -74% | 0% |
A browser automation harness designed for autonomous agents. Maintains a persistent session across tool calls, exposes a CLI subcommand surface, and returns structured JSON the calling agent can parse.
agent-browser open https://example.com
agent-browser screenshot --out /tmp/page.png
agent-browser fill '#email' user@example.comThe browser stays alive between invocations until the session is closed explicitly, so navigation state survives across multiple agent steps.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 4,621 | 2,742 | -41% | 1 | 1 | 0% | 960 | 557 | -42% | 0 | 0 | — |
case-02 | fail→pass | 5,232 | 2,584 | -51% | 1 | 1 | 0% | 974 | 489 | -50% | 0 | 0 | — |
case-03 | fail→pass | 4,356 | 1,910 | -56% | 1 | 1 | 0% | 431 | 426 | -1% | 0 | 0 | — |
case-04 | fail→pass | 3,560 | 4,066 | +14% | 1 | 1 | 0% | 614 | 386 | -37% | 0 | 0 | — |
case-05 | fail→pass | 8,037 | 2,209 | -73% | 1 | 1 | 0% | 1,636 | 431 | -74% | 0 | 0 | — |
case-06 | pass→pass | 9,718 | 3,232 | -67% | 1 | 1 | 0% | 1,825 | 411 | -77% | 0 | 0 | — |
case-07 | fail→pass | 5,489 | 5,275 | -4% | 1 | 1 | 0% | 1,029 | 584 | -43% | 0 | 0 | — |
case-08 | pass→pass | 12,329 | 3,049 | -75% | 1 | 1 | 0% | 2,053 | 615 | -70% | 0 | 0 | — |
case-09 | fail→pass | 14,586 | 2,671 | -82% | 1 | 1 | 0% | 2,586 | 578 | -78% | 0 | 0 | — |
case-10 | pass→pass | 3,860 | 1,412 | -63% | 1 | 1 | 0% | 635 | 299 | -53% | 0 | 0 | — |
case-11 | fail→pass | 6,657 | 3,132 | -53% | 1 | 1 | 0% | 1,320 | 292 | -78% | 0 | 0 | — |
case-12 | fail→pass | 10,209 | 1,616 | -84% | 1 | 1 | 0% | 1,658 | 341 | -79% | 0 | 0 | — |
case-13 | pass→pass | 3,046 | 1,860 | -39% | 1 | 1 | 0% | 555 | 366 | -34% | 0 | 0 | — |
case-14 | fail→pass | 4,339 | 1,702 | -61% | 1 | 1 | 0% | 737 | 395 | -46% | 0 | 0 | — |
case-15 | fail→pass | 4,356 | 1,612 | -63% | 1 | 1 | 0% | 761 | 373 | -51% | 0 | 0 | — |
case-16 | fail→pass | 5,377 | 1,498 | -72% | 1 | 1 | 0% | 1,159 | 317 | -73% | 0 | 0 | — |
case-17 | fail→pass | 4,744 | 6,476 | +37% | 1 | 1 | 0% | 958 | 1,151 | +20% | 0 | 0 | — |
case-18 | fail→pass | 3,594 | 1,727 | -52% | 1 | 1 | 0% | 579 | 294 | -49% | 0 | 0 | — |
case-19 | pass→pass | 7,212 | 3,817 | -47% | 1 | 1 | 0% | 1,210 | 396 | -67% | 0 | 0 | — |
case-20 | pass→pass | 12,161 | 1,470 | -88% | 1 | 1 | 0% | 2,128 | 281 | -87% | 0 | 0 | — |
case-21 | fail→pass | 6,201 | 1,693 | -73% | 1 | 1 | 0% | 1,194 | 388 | -68% | 0 | 0 | — |
case-22 | pass→pass | 6,306 | 4,284 | -32% | 1 | 1 | 0% | 1,260 | 885 | -30% | 0 | 0 | — |
case-23 | pass→pass | 2,248 | 2,671 | +19% | 1 | 1 | 0% | 391 | 499 | +28% | 0 | 0 | — |
case-24 | pass→pass | 2,412 | 2,944 | +22% | 1 | 1 | 0% | 509 | 657 | +29% | 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. 24 cases were attempted. The headline lift of +63 percentage points is the difference between those two pass rates over the 24 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/3/2026 | +50% |
Other measured skills in the registry, with their headline benchmark lift.