Vetting the Blast Radius: The AI Skills Security Index

The Skills Security Index by Tego AI analyzes and scores the security risks of modular instructions used by AI agents. By evaluating capabilities and permissions, it identifies vulnerabilities like prompt injection and data exfiltration across a scale from Pass to Critical. This repository allows developers to understand the security implications of AI skills before deploying them in production environments.

Key Points

AI agent skills represent a new and significant attack surface where instructions can be used to bypass security guardrails.
The index provides a standardized risk ranking framework ranging from 'Pass' to 'Critical' based on the severity of detected instructional risks.
Analysis covers multiple capability dimensions including code execution, web access, file system permissions, and authentication.
The tool identifies specific security deviations such as prompt injection vulnerabilities, credential exposure, and unjustified resource requests.
The primary goal is to help security engineers understand the potential impact and 'blast radius' of a skill before it is integrated into an AI agent.

Sentiment

The community reaction is mixed but leans slightly skeptical. The most visible independent comment is dismissive, viewing the project as marketing for an obvious problem. However, thoughtful counter-arguments emerge around the npm supply-chain analogy and the unique risk of agents acting without human judgment. A minority found immediate practical value. The heavy participation by the project creator adds to the self-promotional perception.

In Agreement

The analogy to npm package security is apt — agent skills are modular capabilities from semi-trusted sources that introduce supply chain risks, and the ecosystem needs trust-building mechanisms similar to what exists for package managers.
AI agents lack the human judgment layer that evaluates consequences before acting, making security vetting of their capabilities more important than for manual terminal commands.
At least one user found practical value in the database, discovering security risks in repos they had already installed without vetting.

Opposed

The concept of cataloging insecure skills is redundant because the underlying problem is just engineers blindly running untrusted code, which is no different from copy-pasting terminal commands — making the database superficial marketing rather than a genuine security contribution.
The project has a self-promotional quality, with the creator actively participating in discussions and the submission reading more like a press release than a community contribution.