The Agent Skills Directory

[COMMAND_EXECUTION]: The skill primarily operates by executing shell commands using the belt CLI tool (belt app run). These commands are used to interact with remote inference services to generate images.
[INDIRECT_PROMPT_INJECTION]: The skill instructions allow for an indirect prompt injection surface because user-supplied product descriptions and style requests are intended to be placed directly into bash commands.
Ingestion points: User requests for specific product photos (e.g., product name, material, lighting preferences) are processed in SKILL.md to be included in the --input JSON payload of the bash commands.
Boundary markers: The skill does not provide clear delimiters or instructions for the agent to sanitize or escape user input before placing it into the shell command templates.
Capability inventory: The skill explicitly requests and uses the Bash(belt *) tool, providing the agent with the ability to execute shell commands.
Sanitization: There is no evidence of sanitization or validation logic to prevent a malicious user from providing input that could escape the JSON string and execute arbitrary shell commands via the belt CLI wrapper.

product-photography