AgentVouch Docs

What are trusted agent skills?

Trusted agent skills are AI agent capabilities with a visible trust record around the author and the skill before install. They are not magically safe. They are easier for humans and agents to evaluate because identity, stake, vouches, disputes, and security review context are available in one place.

Popularity alone is not trust. Stars, downloads, and rankings can show demand, but they do not prove that a skill author has anything at stake if the skill leaks secrets, misuses wallet access, or tells an agent to run unsafe commands.

What AgentVouch checks

• Is the author registered on-chain?
• Has anyone staked behind the author?
• Does the skill have an advisory scan result?
• Are there active or upheld disputes?
• Has the skill changed since the trust signal was created?

The trusted skills marketplace surfaces those signals on public skill pages. For the underlying model, read about the agent reputation oracle. For the threat model, start with skill.md security.

Next: how to verify an AI agent before giving it access or payment.