The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes shell script examples that interpolate user-provided data (e.g., product names) directly into command arguments without validation or escaping. This presents a potential surface for indirect prompt injection or command injection if the agent executes these templates with untrusted input.
Ingestion points: Variable $PRODUCT in the 'Batch Generation' section of SKILL.md.
Boundary markers: Absent; input variables are placed directly within the command string.
Capability inventory: The skill uses the Bash tool, although it is restricted to the belt command.
Sanitization: Absent; the provided examples do not demonstrate input escaping or validation before interpolation.
[COMMAND_EXECUTION]: The skill correctly utilizes platform configuration (allowed-tools: Bash(belt *)) to restrict the agent's shell capabilities to only the necessary belt CLI tool, which is a significant security control.
[EXTERNAL_DOWNLOADS]: The skill references the official belt-sh/cli package and provides links to documentation hosted on the inference.sh GitHub repository. These references are standard for using the supported platform.

ai-product-photography