Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Quick web scanning — discover pages, get snippets, find URLs. For orientation only, not substantive analysis.
.claude/skills/yogsoth-ai-web-search/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 8% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 14% | 0% |
Fast orientation search. Understand what pages and sources exist on a topic. Suitable for quick fact-checking, getting an overview of a landscape, or finding URLs for deeper research via web-research.
Use this when you need to:
This skill returns snippets only. For full-page content analysis, use web-research.
Use whichever web-search MCP is configured — check availability yourself. If more than one is active, try them in the order listed below. This order is just the writing order, not a hard priority — any active provider is equally valid.
Set the result count to ~10 per call (provider-specific parameter named in each Provider Details subsection).
<HARD-GATE> Web search returns snippets only (1-3 sentences per result).
Snippets are NOT authoritative content. They are orientation signals.
PROHIBITED:
REQUIRED:
web-research</HARD-GATE>
Call the configured provider's search tool with ~10 results per call. See Provider Details for exact tool name and parameters.
For each result, present:
Available tools:
| Tool | Purpose | Returns | |------|---------|---------| | brave_web_search | General web search | URL, title, description snippet | | brave_news_search | Recent news articles | URL, title, snippet, date | | brave_video_search | Video content discovery | URL, title, description, duration | | brave_image_search | Image search | URL, title, image properties | | brave_local_search | Local businesses/places | Name, address, rating, hours | | brave_summarizer | AI-generated summary (Pro only) | Summarized text with references |
Key parameters:
brave_web_search:query (required): search terms, max 400 charscount: 1-20 results (default 10)offset: pagination, 0-9 (default 0)freshness: pd (24h), pw (7d), pm (31d), py (365d), or YYYY-MM-DDtoYYYY-MM-DDsafesearch: off / moderate / strict (default moderate)search_lang: language code (default en)country: 2-letter country code (default US)result_filter: array of types to include (web, news, videos, etc.)brave_news_search: same core parameters, better for time-sensitive queriesbrave_video_search: returns duration and thumbnail_urlbrave_image_search: count 1-200 (default 50)brave_local_search: for "near me" queries, returns business detailsbrave_summarizer: requires Pro plan, must first run brave_web_search with summary=trueExamples:
Quick fact check:
brave_web_search(query="Claude 3.5 Sonnet release date", count=5)News monitoring:
brave_news_search(query="transformer architecture breakthroughs", freshness="pm", count=10)Landscape scan:
brave_web_search(query="MCP server frameworks comparison 2025", count=15)Available tools:
| Tool | Purpose | Returns | |------|---------|---------| | tavily_search | LLM-optimized web search | URL, title, content snippet, score |
Key parameters:
tavily_search:query (required): search termsmax_results: number of results (default 10)search_depth: basic or advanced (default basic)include_domains: restrict to specific domainsexclude_domains: exclude specific domainsExamples:
Quick fact check:
tavily_search(query="Claude 3.5 Sonnet release date", max_results=5)Landscape scan:
tavily_search(query="MCP server frameworks comparison 2025", max_results=10)Keyless by default — the keenable MCP server needs no API key.
Available tools:
| Tool | Purpose | Returns | |------|---------|---------| | search_web_pages | Web search over the Keenable index | title, URL, description snippet | | fetch_page_content | Fetch a URL as clean markdown | title, URL, full page content |
Key parameters:
search_web_pages:query (required): search termsfetch_page_content:url (required): page to fetch as markdownExamples:
Quick fact check:
search_web_pages(query="Claude 3.5 Sonnet release date")Landscape scan then deep read:
search_web_pages(query="MCP server frameworks comparison 2025")
fetch_page_content(url="https://...")<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | web-research | Deep web research — fetches full page content for analysis. Snippets alone are PROHIBITED for conclusions. |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 15,125 | 20,119 | +33% | 1 | 1 | 0% | 1,550 | 2,100 | +35% | 0 | 0 | — |
case-02 | fail→fail | 16,382 | 18,388 | +12% | 1 | 1 | 0% | 252 | 2,143 | +750% | 0 | 0 | — |
case-03 | fail→fail | 13,932 | 17,135 | +23% | 1 | 1 | 0% | 423 | 2,269 | +436% | 0 | 0 | — |
case-04 | fail→fail | 4,456 | 8,143 | +83% | 1 | 1 | 0% | 671 | 2,213 | +230% | 0 | 0 | — |
case-05 | pass→pass | 8,110 | 7,536 | -7% | 1 | 1 | 0% | 600 | 2,111 | +252% | 0 | 0 | — |
case-06 | pass→pass | 8,298 | 7,396 | -11% | 1 | 1 | 0% | 615 | 2,044 | +232% | 0 | 0 | — |
case-07 | fail→pass | 16,923 | 7,322 | -57% | 1 | 1 | 0% | 1,794 | 1,947 | +9% | 0 | 0 | — |
case-08 | fail→pass | 17,878 | 3,378 | -81% | 1 | 1 | 0% | 1,955 | 2,242 | +15% | 0 | 0 | — |
case-09 | fail→pass | 16,718 | 7,659 | -54% | 1 | 1 | 0% | 1,933 | 2,084 | +8% | 0 | 0 | — |
case-10 | pass→pass | 9,003 | 10,072 | +12% | 1 | 1 | 0% | 1,596 | 2,452 | +54% | 0 | 0 | — |
case-11 | pass→pass | 7,726 | 6,839 | -11% | 1 | 1 | 0% | 446 | 1,982 | +344% | 0 | 0 | — |
case-12 | pass→pass | 10,131 | 8,458 | -17% | 1 | 1 | 0% | 1,938 | 2,247 | +16% | 0 | 0 | — |
case-13 | pass→pass | 9,737 | 9,511 | -2% | 1 | 1 | 0% | 750 | 2,421 | +223% | 0 | 0 | — |
case-14 | pass→pass | 4,286 | 8,189 | +91% | 1 | 1 | 0% | 727 | 2,071 | +185% | 0 | 0 | — |
case-15 | fail→pass | 15,599 | 7,188 | -54% | 1 | 1 | 0% | 1,690 | 1,962 | +16% | 0 | 0 | — |
case-16 | fail→pass | 16,179 | 7,345 | -55% | 1 | 1 | 0% | 1,761 | 2,010 | +14% | 0 | 0 | — |
case-17 | fail→pass | 20,881 | 1,955 | -91% | 1 | 1 | 0% | 2,241 | 1,934 | -14% | 0 | 0 | — |
case-18 | fail→pass | 28,017 | 7,340 | -74% | 1 | 1 | 0% | 4,179 | 1,918 | -54% | 0 | 0 | — |
case-19 | fail→pass | 14,110 | 3,518 | -75% | 1 | 1 | 0% | 1,376 | 2,178 | +58% | 0 | 0 | — |
case-20 | pass→pass | 8,043 | 6,402 | -20% | 1 | 1 | 0% | 538 | 1,879 | +249% | 0 | 0 | — |
case-21 | pass→pass | 7,996 | 6,663 | -17% | 1 | 1 | 0% | 507 | 1,940 | +283% | 0 | 0 | — |
case-22 | pass→pass | 11,790 | 7,830 | -34% | 1 | 1 | 0% | 1,289 | 2,277 | +77% | 0 | 0 | — |
case-23 | pass→pass | 11,593 | 7,680 | -34% | 1 | 1 | 0% | 2,088 | 2,149 | +3% | 0 | 0 | — |
case-24 | fail→pass | 51,289 | 22,474 | -56% | 1 | 1 | 0% | 8,231 | 4,940 | -40% | 0 | 0 | — |
case-25 | fail→pass | 8,890 | 13,837 | +56% | 1 | 1 | 0% | 612 | 2,365 | +286% | 0 | 0 | — |
case-26 | fail→pass | 6,771 | 10,305 | +52% | 1 | 1 | 0% | 1,109 | 2,552 | +130% | 0 | 0 | — |
case-27 | pass→pass | 9,732 | 7,314 | -25% | 1 | 1 | 0% | 871 | 2,089 | +140% | 0 | 0 | — |
case-28 | pass→pass | 17,749 | 2,965 | -83% | 1 | 1 | 0% | 2,048 | 2,167 | +6% | 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. 28 cases were attempted, and 25 counted toward the lift figure. The other 3 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 +39 percentage points is the difference between those two pass rates over the 25 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.