Researchers found 341 malicious skills in one AI agent marketplace
jayesh_ahire1 · x · 2026-07-29
A security write-up says the team found 341 malicious skills in one agent-skill marketplace, roughly one in eight.
The post argues that the real barrier is not just malicious code, but how easy it is to publish dangerous agent skills: a GitHub account older than one week was enough to get listed. The linked article frames agent risk as a construction and governance problem, not only an attack problem, and uses the "butter robot" analogy to stress how purpose, scope, capability, and human accountability can be aligned in a well-governed agent.
Related event: Hundreds of Malicious Skills Found in AI Agent Marketplaces(2 posts)→
More from coding & agent
- Together AI schedules a July 30 deep dive on Kimi K3 architecture and production inference — togethercompute · 2026-07-29
- Command Center builds a review layer for oversized diffs from coding agents — jimmykoppel · 2026-07-29
- YC startup hwintelligence launches Wave, a waveform debugger with a verification agent — ycombinator · 2026-07-29
- Claude turns a 30,000-line localization app into a single prompt — draginol · 2026-07-29
- Two coding agents found and fixed the same Bun bug overnight — Delicious-Flan88 · 2026-07-29
- MoonshotAI’s FlashKDA commit hints at a hybrid attention mainline model — peterjliu · 2026-07-29