Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Browse the web, take screenshots, interact with pages, fill forms. Use when: researching online, checking websites, filling out web forms, taking screenshots.
.claude/skills/outworked-web-browser/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-18 | ✓→✓ | ＝ Same ✓ | 8% | 0% |
| case-20 | ✓→✓ | ＝ Same ✓ | 39% | 0% |
You can browse the web using the browse: tools. A managed browser window handles navigation and interaction.
url (string).selector (string).<input> and <textarea> elements. Params: selector (string), value (string).selector to auto-focus. Params: text (string, required), selector (string, optional), clearFirst (boolean, optional).script (string).url (string, optional), message (string, optional).url (string, optional), message (string, optional).browse:navigate — the response includes all interactive elements with selectorsbrowse:screenshot if you need to visually verify the page statebrowse:snapshot to refresh the interactive element list without navigatingbrowse:evaluate to find elements — the interactive snapshot gives you everything[aria-label="Like"])browse:login to let the user sign inbrowse:fill for standard form inputs (<input>, <textarea>, <select>). Use browse:type for rich text editors, contentEditable fields, or any site where browse:fill doesn't work (Twitter/X, Notion, Google Docs, etc.)| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 4,030 | 5,737 | +42% | 1 | 1 | 0% | 524 | 1,106 | +111% | 0 | 0 | — |
case-01 | fail→fail | 3,720 | 4,422 | +19% | 1 | 1 | 0% | 627 | 918 | +46% | 0 | 0 | — |
case-03 | fail→fail | 3,065 | 3,684 | +20% | 1 | 1 | 0% | 416 | 867 | +108% | 0 | 0 | — |
case-04 | fail→fail | 9,556 | 4,389 | -54% | 1 | 1 | 0% | 1,634 | 987 | -40% | 0 | 0 | — |
case-05 | fail→pass | 4,792 | 1,920 | -60% | 1 | 1 | 0% | 730 | 1,036 | +42% | 0 | 0 | — |
case-06 | fail→fail | 3,007 | 4,115 | +37% | 1 | 1 | 0% | 441 | 925 | +110% | 0 | 0 | — |
case-07 | fail→pass | 6,484 | 2,134 | -67% | 1 | 1 | 0% | 970 | 1,100 | +13% | 0 | 0 | — |
case-08 | fail→fail | 7,281 | 4,022 | -45% | 1 | 1 | 0% | 1,268 | 887 | -30% | 0 | 0 | — |
case-09 | fail→fail | 3,602 | 5,225 | +45% | 1 | 1 | 0% | 486 | 1,042 | +114% | 0 | 0 | — |
case-10 | fail→fail | 5,277 | 4,807 | -9% | 1 | 1 | 0% | 884 | 977 | +11% | 0 | 0 | — |
case-11 | fail→fail | 4,321 | 3,429 | -21% | 1 | 1 | 0% | 725 | 835 | +15% | 0 | 0 | — |
case-12 | fail→fail | 12,955 | 15,853 | +22% | 1 | 1 | 0% | 352 | 1,563 | +344% | 0 | 0 | — |
case-13 | fail→fail | 7,249 | 4,666 | -36% | 1 | 1 | 0% | 1,363 | 910 | -33% | 0 | 0 | — |
case-14 | fail→fail | 11,653 | 5,016 | -57% | 1 | 1 | 0% | 2,313 | 1,031 | -55% | 0 | 0 | — |
case-15 | fail→fail | 9,902 | 4,772 | -52% | 1 | 1 | 0% | 1,736 | 986 | -43% | 0 | 0 | — |
case-16 | fail→fail | 5,189 | 3,757 | -28% | 1 | 1 | 0% | 895 | 821 | -8% | 0 | 0 | — |
case-17 | fail→pass | 9,281 | 4,992 | -46% | 1 | 1 | 0% | 1,513 | 1,495 | -1% | 0 | 0 | — |
case-18 | pass→pass | 4,810 | 1,630 | -66% | 1 | 1 | 0% | 918 | 996 | +8% | 0 | 0 | — |
case-19 | fail→fail | 6,072 | 4,890 | -19% | 1 | 1 | 0% | 943 | 1,014 | +8% | 0 | 0 | — |
case-20 | pass→pass | 12,437 | 13,447 | +8% | 1 | 1 | 0% | 2,525 | 3,507 | +39% | 0 | 0 | — |
case-21 | pass→pass | 6,495 | 7,898 | +22% | 1 | 1 | 0% | 1,361 | 1,797 | +32% | 0 | 0 | — |
case-22 | pass→pass | 4,272 | 2,399 | -44% | 1 | 1 | 0% | 685 | 1,101 | +61% | 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, and 8 counted toward the lift figure. The other 14 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +14 percentage points is the difference between those two pass rates over the 8 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.