The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to install the belt CLI tool and references configuration files from the inference-sh GitHub repository. These resources trace back to the official infrastructure of the service provider described in the skill.
[COMMAND_EXECUTION]: The skill interacts with the inference platform using the belt command-line utility. The execution environment is restricted via the allowed-tools field to only permit the belt command, which adheres to the principle of least privilege and prevents arbitrary shell execution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests natural language prompts and image URLs provided by the user to facilitate image generation.
Ingestion points: The prompt, images, and mask parameters within the tool's JSON input.
Boundary markers: External data is encapsulated within a structured JSON object passed to the CLI tool.
Capability inventory: The skill's capabilities are confined to the specific operations supported by the belt application.
Sanitization: Inputs are treated as data parameters for the remote image generation model.

gpt-image