Data study · v1, October 2026
How Claude finds agent skills it does not know, in Claude Code
The short answer
When Claude is asked for an agent skill it does not know, it searches the web: all 18 search runs on little-known skills searched, on three Claude models. The page that reached the answer was usually one page per skill on a skill directory: 13 of 18 answers linked a third-party skill directory, and 17 of 18 linked the skill's own GitHub repository. For famous skills, Claude went to the publisher's own repository instead. The counts in this answer come from 93 runs on , through Claude Code and the API; the page-opening and first-pick counts further down pool these runs with an earlier test, and say so.
Key takeaways
- All 18 search runs on little-known skills searched the web, Sonnet 5.5 included, which skipped search on most other questions.
- 13 of 18 answers linked a third-party skill directory in the answer body; skills.lc was linked most, 6 times, then mcpmarket.com 4 times.
- For broad directory questions, Claude named the leader from memory too: There's An AI For That was named in 27 of 27 search runs and 9 of 9 memory-only runs.
- A precise attribute let new sites in: tooldirectory.ai, with a number in its title, entered 8 of 9 answers to the widest-categories question.
- Across this test and an earlier one (147 runs), Claude opened a page in only 55 of 489 cases where a domain was cited; most answers were written from search result titles and memory.
- The head is stable, the order is not: over 15 runs on MCP servers for web research, pooled across this test and an earlier one, the first pick split Tavily 6, Brave 4, Exa 3, Fetch 2.
When Claude is asked for an agent skill it does not know, it searches, and the page that reaches the answer is usually one page per skill on a skill directory. In 93 runs on three Claude models on , all 18 search runs on little-known skills searched the web (18 of 18), and 13 of 18 answers linked a third-party skill directory in the answer itself. For famous skills, Claude went to the publisher's own repository instead.
How did we run the test?
| Date | |
|---|---|
| Models | claude-opus-5-5, claude-sonnet-5-5, claude-fable-5-1, through Claude Code 2.1.286 (API, web search on) |
| Setup | An empty folder, no user settings, no memory, only the web search and web fetch tools |
| Questions | 9 plain questions, no instructions: 3 about AI tool directories, 3 about little-known agent skills, Instagram posting, new coding tools, MCP servers for web research |
| Runs | 93: 2 to 3 search runs and 1 memory-only run per model and question, plus extra repeats for stability |
| What we counted | Links in the answer body. Claude Code's search tool also appends a Sources list to every answer, which we did not count as a citation |
All questions are listed below.
Results
Little-known skills go to skill directories. All 18 search runs searched (Sonnet 5.5 too, which skipped search on most other questions).
What did Claude do on the 18 searches for little-known skills?
- Searched the web18 of 18
- Linked the skill's own GitHub repository17 of 18
- Linked a third-party skill directory13 of 18
18 search runs on the three skill questions, on Claude Opus 5.5, Sonnet 5.5 and Fable 5.1, . Links counted in the answer body only, not in the Sources list Claude Code appends.
Source: CookiesForAI, v1, October 2026, 18 runs
Reuse this chart
Free to reuse with credit (CC BY 4.0). Download SVG
<figure><img src="https://cookiesforai.app/charts/how-claude-finds-agent-skills-october-2026--skill-searches.svg" alt="What did Claude do on the 18 searches for little-known skills: Searched the web 18 of 18, Linked the skill's own GitHub repository 17 of 18, Linked a third-party skill directory 13 of 18."><figcaption>Source: CookiesForAI, What did Claude do on the 18 searches for little-known skills?, https://cookiesforai.app/research/how-claude-finds-agent-skills-october-2026</figcaption></figure>The answer bodies linked: skills.lc 6 times, mcpmarket.com 4, explainx.ai 3, agent-skills.md 3, agentskill.sh 2, lobehub.com 2, crossaitools.com 2, skillstore.io 2, and six others once. The answers linked the skill's own GitHub repository in 17 of 18 runs. The pages that won had one page per skill with the skill's name in the address.
Which skill directories did the answers link?
- skills.lc6
- mcpmarket.com4
- explainx.ai3
- agent-skills.md3
- agentskill.sh2
- lobehub.com2
- crossaitools.com2
- skillstore.io2
How often each directory was linked in the answer bodies of the 18 skill search runs; six other directories were linked once each.
Source: CookiesForAI, v1, October 2026
Reuse this chart
Free to reuse with credit (CC BY 4.0). Download SVG
<figure><img src="https://cookiesforai.app/charts/how-claude-finds-agent-skills-october-2026--directory-links.svg" alt="Which skill directories did the answers link: skills.lc 6 links, mcpmarket.com 4 links, explainx.ai 3 links, agent-skills.md 3 links, agentskill.sh 2 links, lobehub.com 2 links, crossaitools.com 2 links, skillstore.io 2 links."><figcaption>Source: CookiesForAI, Which skill directories did the answers link?, https://cookiesforai.app/research/how-claude-finds-agent-skills-october-2026</figcaption></figure>Claude searched in the user's words plus "SKILL.md". Typical query: "agent skill Terraform plan review SKILL.md", often with "Claude Code" added.
Broad directory questions were decided from memory. There's An AI For That was named in 27 of 27 search runs and 9 of 9 memory-only runs of the three directory questions, and Futurepedia in all 27 search runs.
A precise attribute let new sites in. "Which AI tools directory has the widest range of categories?" made all 9 runs search, and two sites no memory run named entered the answers: tooldirectory.ai in 8 of 9, with a number in its title, and aicentralresources.com in 7 of 9.
Pages were rarely opened. Across this test and an earlier one (147 runs), Claude opened a page in 55 of 489 cases where a domain was cited; most answers were written from search result titles and memory. Titles shaped as "best" or "top" lists reached the answer body 4% of the time, against 20% for other titles.
The head is stable; the order is not. Over 15 runs on MCP servers for web research, pooled across this test and an earlier one, Brave Search, Exa, Tavily, Firecrawl, Fetch and Playwright were named every time, but the first pick split Tavily 6 of 15, Brave 4 of 15, Exa 3 of 15, Fetch 2 of 15.
Which MCP server did Claude name first for web research?
- Tavily6 of 15
- Brave4 of 15
- Exa3 of 15
- Fetch2 of 15
The first pick in 15 search runs, pooled across this test and an earlier one, on "Which MCP servers should I connect to my AI agent for web research?". Brave Search, Exa, Tavily, Firecrawl, Fetch and Playwright were named in every run.
Source: CookiesForAI, v1, October 2026, 15 runs
Reuse this chart
Free to reuse with credit (CC BY 4.0). Download SVG
<figure><img src="https://cookiesforai.app/charts/how-claude-finds-agent-skills-october-2026--first-pick.svg" alt="Which MCP server did Claude name first for web research: Tavily 6 of 15, Brave 4 of 15, Exa 3 of 15, Fetch 2 of 15."><figcaption>Source: CookiesForAI, Which MCP server did Claude name first for web research?, https://cookiesforai.app/research/how-claude-finds-agent-skills-october-2026</figcaption></figure>YouTube barely appears. 2 of 1,483 search results Claude was shown were YouTube videos, and none reached an answer.
What does this mean for a little-known skill?
For an agent skill that a model does not already know, the way into an answer is a page about that one skill, named with the skill's name and the task words people use, on a site the search engine already shows. This is why CookiesForAI has one page per skill with its name in the address.
What are the limits?
Claude only, through Claude Code and the API; the claude.ai app may cite differently. Small samples: 2 or 3 search runs per model and question. Results hold for these three models on this date; newer models may behave differently. The same questions on ChatGPT, Gemini, Perplexity, Google and Grok are in our six-engine study.
The questions we asked
AI tool directory questions (3)
- What is the best AI tools directory for finding new AI tools?
- Which AI tools directory includes user reviews and ratings?
- Which AI tools directory has the widest range of categories?
Skill questions (3)
- Is there an agent skill that helps my coding agent review Terraform plans?
- What agent skills exist for making videos with code?
- Find me an agent skill for writing database migrations safely.
Other questions (3)
- How do I post to Instagram from Claude?
- What new AI coding tools launched this month that are worth trying?
- Which MCP servers should I connect to my AI agent for web research?
Data and how to cite
Download the chart data: CSV, JSON. Free to reuse with credit (CC BY 4.0).
CookiesForAI, "How Claude finds agent skills it does not know, in Claude Code", v1, October 2026, https://cookiesforai.app/research/how-claude-finds-agent-skills-october-2026CookiesForAI lists agent skills, MCP servers and apps, so our own pages can appear in these answers; when they do, we count them like any other site.
Changes to this page
- : corrected the summary: it said every question about a little-known skill made Claude search; the count is all 18 search runs on those questions.
- : title now says the runs were in Claude Code; short answer, takeaways, charts and the question list added; corrected: six other directories were linked once, not seven; the first-pick counts now say they pool 15 runs across this test and an earlier one.




