The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to install Ollama using its official setup script via curl -fsSL https://ollama.ai/install.sh | sh. While this pattern involves piping remote code to a shell, the source is a well-known service for LLM deployment.
[INDIRECT_PROMPT_INJECTION]: The inference server implementations in SKILL.md and scripts/inference_server.py create a surface for indirect prompt injection by processing untrusted user input.
Ingestion points: The prompt and messages fields in the FastAPI endpoints.
Boundary markers: Standard model-specific tags (e.g., [INST]) are used for formatting, but no specialized security delimiters are present to isolate user instructions.
Capability inventory: The system is limited to text generation; it does not expose dangerous capabilities like arbitrary shell execution or file system writes based on the prompt content.
Sanitization: The example code does not include input sanitization or output filtering.
[COMMAND_EXECUTION]: The skill includes standard commands for managing deployments, such as pip install vllm for package management and docker run for container orchestration, which are appropriate for its stated purpose.

model-deployment