The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands through the Bash tool to interact with the belt CLI for model inference and account management. This execution is limited to the belt command as defined in the skill's allowed-tools configuration.
[INDIRECT_PROMPT_INJECTION]: The skill processes external inputs such as URLs for images and audio, as well as text scripts, which are passed as arguments to CLI commands. This represents a vulnerability surface for indirect prompt injection, where malicious instructions could be embedded in the external content.
Ingestion points: Example CLI commands in SKILL.md demonstrate passing URLs and text strings via the --input flag.
Boundary markers: JSON formatting is used for input parameters to provide data separation.
Capability inventory: Shell access is restricted to the belt command, limiting the impact of potential injection.
Sanitization: The skill relies on the underlying belt CLI tool for validation and handling of external parameters.
[EXTERNAL_DOWNLOADS]: The skill references setup documentation hosted on the official GitHub repository for the inference-sh service, which is used for configuring the required CLI tool.

ai-avatar-video