What permissions do AI agents have when a skill is active?

When a skill is active, the agent has the same permissions as the user, allowing it to read and write files, run shell commands, access environment variables, make network requests, and interact with databases.

How can I evaluate skills for security before installing them?

Before installing, read the `SKILL.md` and any scripts for suspicious commands, consider the source's reputation, and prioritize skills from reviewed marketplaces like Agensi, which conduct security scans.

What kind of suspicious commands should I look for in a skill?

Look for commands that access environment variables, make network requests to unknown URLs, download and execute remote scripts, or modify files outside the intended project directory.

What security checks does Agensi perform on skills?

Agensi performs an automated 8-point security scan and manual review, checking for dangerous command patterns, hardcoded secrets, obfuscation, environment variable harvesting, and prompt injection attempts.

What steps should I take to write secure SKILLs?

To write secure skills, avoid hardcoding secrets, scope file access within the project, explicitly define commands, use `disable-model-invocation` for risky actions, and document all accessed resources.

Why is it important to avoid hardcoding secrets in skill development?

Hardcoding secrets like API keys or tokens in your `SKILL.md` is a security risk; instead, reference environment variables to allow the agent to access them securely at runtime.

When should 'disable-model-invocation: true' be used in skills?

Skills that perform risky actions such as deploying code, sending messages, or making irreversible changes should always have `disable-model-invocation: true` to ensure they only run with explicit user invocation.

Now open worldwide. Creators in every country can onboard and sell agent skills on Agensi -> Start selling

Skill md Security Best Practices