Unverified50% confidenceBenchmarkExact time
In an audit of nearly 4,000 skills on Claude Hub, more than one in eight had critical-level issues, and 76 malicious payloads were discovered.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/14/2026
First Seen
Valid until: 11/12/2026
Sources
AI Agent安全防护：Snyk三大防线防止删库跑路
Related Claims
UnverifiedSurvey data shows that 341 out of 2,857 Skills contained toxic or harmful content, accounting for more than one tenth.65% similarUnverifiedEmpirical data shows that with more than 60 Skills installed, only about 6% of the context window is occupied at startup.63% similarUnverifiedAnthropic's "needle test" benchmark shows that Claude achieves a key information retrieval accuracy of over 99% within a 200,000-token range.62% similarUnverifiedA 2018 survey of 5,000 programmers showed that approximately 58% of respondents had experienced impostor syndrome to varying degrees59% similarUnverifiedExpert users triggered more than twice as many Claude actions per instruction as novices, with five times the output volume (12 actions vs. 5, and 3,200 words vs. 600 words)59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/745884API
curl https://kongchang.com/api/v1/knowledge/claims/745884MCP
get_claim(id=745884)