The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data through the prompt, source, and imageRef inputs which are used to generate or edit images via external AI models.
Ingestion points: Untrusted data enters the agent context via the prompt string and imageRef/source asset IDs defined in the SKILL.md instructions.
Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" markers when interpolating user input into the tool parameters.
Capability inventory: The skill uses model_run to call external APIs, asset_download to write files to the local system, and upload_asset to read files from the local system.
Sanitization: There is no mention of sanitizing or validating user-provided strings before they are passed to the model_run tool.
[COMMAND_EXECUTION]: The instructions suggest the use of npx skills add scenario-labs/skills --skill <name> to the agent as a way to resolve missing dependencies. This involves the execution of shell commands that interact with external package managers and code repositories associated with the vendor.

scenario-luma-image