The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages the execution of shell commands, including binary version checks and tool installation commands. Safety is maintained through mandatory user oversight and a requirement for the agent to verify commands against official documentation before proposing them.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external curriculum files, which serves as an ingestion point for potentially untrusted content that could attempt to influence the agent's behavior. 1. Ingestion points: The skill reads from local curriculum files (../companion/references/course.md) and fetches public course materials. 2. Boundary markers: Explicit instructions are provided to treat curriculum content as data only, not as an authority to execute commands. 3. Capability inventory: The agent can execute terminal commands and modify local records like tools-learned.md. 4. Sanitization: Instructions require the agent to inspect all commands for malicious intent or requests for secrets before presenting them to the user.
[EXTERNAL_DOWNLOADS]: The skill assists in downloading and installing software from external repositories based on the course toolkit, relying on the user to approve specific sources and commands after the agent verifies their legitimacy against official documentation.

lab-setup