Unverified50% confidenceFactExact time
The test data is based on the October 17 snapshot of skills.sh.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/19/2026
First Seen
Valid until: 11/2/2026
Sources
Related Entities
Related Claims
UnverifiedThe research tested 7 skills and 4 victim models.67% similarUnverifiedRecommendations for tool evaluation: use deterministic testing as much as possible for what can be deterministically tested, and then use LLM to evaluate touchpoints with agents.67% similarUnverifiedSkill Spectre detects suspicious instructions and abnormal data flows, verifies whether the access permissions requested by skills match their functionality, and provides risk assessment factors.66% similarUnverifiedThe research team released an accompanying benchmark dataset, SkillCascade-Bench.66% similarUnverifiedBenchmark evaluation methods combine automated metrics (CLIP-Score, LPIPS, SSIM) with human evaluation.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1065258API
curl https://kongchang.com/api/v1/knowledge/claims/1065258MCP
get_claim(id=1065258)