The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the belt CLI tool via the Bash tool to execute model inference and manage platform state (e.g., belt login, belt app run).
[EXTERNAL_DOWNLOADS]: The documentation references and provides instructions for installing the belt CLI and additional skills from the inference-sh organization on GitHub (github.com/inference-sh).
[INDIRECT_PROMPT_INJECTION]: The skill provides prompt templates designed to ingest and process untrusted external content, which is a common vulnerability surface.
Ingestion points: Placeholders in shell commands such as [article text] and [code] in SKILL.md.
Boundary markers: The templates lack explicit boundary markers or instructions to ignore embedded commands within the processed data.
Capability inventory: The skill uses the belt tool to communicate with various LLMs and image/video models.
Sanitization: No sanitization or validation of the input data is provided or recommended in the templates.

prompt-engineering