← Back to digest
AI Agents

Skill registry trust study

What's this about?

Researchers surveyed agent-skill registries and report 77.86% of skills had no stars or comments. They found widespread privilege-related evidence in readable skills and substantial disagreement among scanners. The counts describe the sampled registry, not every skill ecosystem.

What exactly is a skill registry?

A catalog of instruction packages that tells an agent how to perform specialised work. Since instructions may request tools or permissions, downloading a skill is closer to adding software than reading a harmless article.

What's the community saying?

No substantial 24-hour community thread was found; this is an author-reported preprint.
  1. Absence of feedback is not evidence of safety.
  2. Scanner disagreement limits simple safe/unsafe badges.
  3. Review permissions, source and behaviour before running a skill.
← The verification bottleneckFyxer's specialist-model assistant →