The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-supplied prompts and pass them to external Large Language Models.
Ingestion points: User input is provided via the --input flag in commands like belt app run within SKILL.md.
Boundary markers: Data is structured using JSON objects (e.g., {"prompt": "..."}).
Capability inventory: The skill utilizes the Bash tool to execute CLI commands and interact with network-based LLM providers.
Sanitization: The skill relies on the target LLM providers (e.g., Anthropic, Google) and the OpenRouter gateway for content safety filtering.
[COMMAND_EXECUTION]: The skill uses the Bash tool restricted to the belt command to interact with the inference.sh platform.
Evidence: The frontmatter contains allowed-tools: Bash(belt *), and the body provides examples for belt login, belt app run, and belt app search.
[EXTERNAL_DOWNLOADS]: The skill provides documentation links to external resources for setting up the necessary CLI environment.
Evidence: References https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md for installation instructions.

llm-models