The Agent Skills Directory

[Instructional Code and Command Execution]: The skill provides lessons on enabling shell access and JavaScript interpretation (using QuickJS). These features are the primary subject matter of the course, and the content includes detailed warnings about security risks and instructions on using sandboxes and Human-in-the-Loop (HITL) approvals to mitigate them.
[Trusted External Downloads]: The instructions guide users to install development tools and clone repository resources from well-known and established sources, such as GitHub and the official sites for the uv package manager. These references are standard for the intended development environment setup.
[Credential Management Guidance]: The tutor is specifically instructed never to ask users for API keys or secrets in the chat. Instead, it teaches users the safe practice of managing credentials in local environment files that are not shared with the agent's context.
[Data Sanitization Practices]: In examples involving the generation of HTML content from external data, the skill demonstrates the use of sanitization libraries like nh3 to prevent injection vulnerabilities.

lca-deepagents