The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install external components via npx skills add and provides links to installation instructions hosted on a public GitHub repository (inference-sh/skills). This involves fetching and executing third-party code to set up the necessary environment.\n- [COMMAND_EXECUTION]: The skill requires the use of the belt CLI tool to perform login and model execution tasks. These operations involve shell commands that interact with the local system and remote services.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-provided strings as prompts for language models. This creates a surface where untrusted data could contain instructions that influence the agent's behavior during interaction with the LLMs.\n
Ingestion points: Prompts and system instructions passed through the --input flag or via JSON files in the SKILL.md examples.\n
Boundary markers: None; the content is passed directly to the model without specific delimiters or warnings to ignore embedded instructions.\n
Capability inventory: The skill can execute CLI commands (belt) and requires network access to communicate with model endpoints.\n
Sanitization: No sanitization or validation of the prompt content is specified in the instruction examples.

llm-models