ToxicSkills e ClawHavoc — A Crise de Segurança nas Skills de Agentes (2026)
Uma auditoria de segurança em 22.511 skills de agentes encontrou 140.963 problemas. A pesquisa ToxicSkills da Snyk identificou injeção de prompt em 36% das skills. O que os desenvolvedores precisam saber.

Uma auditoria de segurança em 22.511 skills de agentes encontrou 140.963 problemas. A pesquisa ToxicSkills da Snyk identificou injeção de prompt em 36% das skills. O que os desenvolvedores precisam saber.
- Author: Shahrukh — Creator of PromptSpace, AI researcher & prompt engineer since 2024. 159+ articles published.
- Methodology: Claims in this article are based on hands-on testing with live AI models, publicly available benchmarks, and official model documentation.
- Last tested: Content reviewed and verified against current model versions as of the publication date above.
- Sources: Official model docs, published research, and curated community examples. Links open in context where available.
- Updates: PromptSpace updates articles when models change significantly. Check the "Updated" date in the header for recency.
Written by Shahrukh
Creator of PromptSpace · AI Researcher & Prompt Engineer
Building the largest free AI prompt library with 4,000+ prompts. Covering AI image generation, prompt engineering, and tool comparisons since 2024. 159+ articles published.