The Agent Skills Directory

[SAFE]: The skill provides legitimate architectural guidance, code examples, and security checklists for building AI/ML applications. Content is educational and matches its stated purpose.
[PROMPT_INJECTION]: The skill contains phrases like 'ignore instructions' and 'reveal system prompt' within references/evaluation.md. These are explicitly used as test cases for the promptfoo evaluation framework to verify the robustness of user-built applications against attacks, rather than attempts to subvert the agent's own behavior.
[EXTERNAL_DOWNLOADS]: Includes instructions to download and install official tools from well-known services, such as Ollama and the llama.cpp repository. These operations target established technology providers and are standard for the described development tasks.
[COMMAND_EXECUTION]: Provides standard shell commands for compiling software from source (llama.cpp) and managing system services (systemd). These are implemented following security best practices, such as recommending against passing API keys via command-line arguments.
[DATA_EXPOSURE]: The skill correctly emphasizes the use of environment variables and secret managers for credential management, with no hardcoded secrets or unsafe data handling practices detected.

ai-ml