The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to construct shell commands using unvalidated user input (e.g., the --prompt "{description}" and --reference flags). This design lacks explicit sanitization requirements, creating a risk where a user could inject malicious shell commands by including metacharacters (like backticks or semicolons) in their image description or reference paths.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its ingestion of untrusted external content.
Ingestion points: User-supplied text prompts and reference image URLs/local paths are consumed in the SKILL.md interaction flow.
Boundary markers: Absent. There are no delimiters or specific warnings provided to the agent to treat the embedded instructions within the prompt as data rather than commands.
Capability inventory: The skill utilizes the listenhub CLI tool, which possesses network access and file system write capabilities.
Sanitization: Absent. No sanitization or validation logic is defined for the user-provided inputs prior to their use in a shell execution context.

image-gen