The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches installation guidelines and documentation from the inference-sh GitHub repository. It also refers to the belt-sh/cli skill which is installed via npx. These are standard external resources required to interact with the inference service provided by the vendor.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, which constitutes a potential attack surface where malicious instructions could be embedded in prompts or images.
Ingestion points: The prompt, images (URLs), and mask (URL) parameters in SKILL.md are used to pass data to the model.
Boundary markers: No explicit delimiters or specific 'ignore instructions' warnings are defined to wrap user-provided inputs when passed to the model.
Capability inventory: The skill has the capability to execute shell commands via the belt CLI tool using Bash(belt *) to perform image generation and editing.
Sanitization: The skill does not implement custom sanitization for these inputs, relying instead on the safety guardrails of the underlying gpt-image-2 model and the belt CLI tool.

gpt-image