The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs a shell command to execute scripts/generate_image.py where the --prompt argument is derived from untrusted user-supplied story descriptions. This pattern is vulnerable to command injection if the user provides input containing shell metacharacters (e.g., semicolons, backticks, or pipes).
[COMMAND_EXECUTION]: The skill hardcodes absolute file paths to a specific user's home directory (/Users/ugreen/Documents/obsidian/09image/) for saving generated images. This reveals local system structure and can lead to unauthorized file creation in sensitive user folders.
[EXTERNAL_DOWNLOADS]: The skill's Python utility fetches data from the Google Gemini API at https://generativelanguage.googleapis.com to perform image generation.
[PROMPT_INJECTION]: The skill processes untrusted user stories that are used for both internal model prompting and command-line execution, creating an indirect prompt injection surface.
Ingestion points: User story descriptions input during Phase 1 in SKILL.md.
Boundary markers: Absent; there are no delimiters or instructions provided to the agent to treat the user story as untrusted data.
Capability inventory: The skill possesses file writing and network communication capabilities through its included Python script.
Sanitization: None; there is no evidence of escaping or validation performed on the story text before it is interpolated into shell commands.

space-storyboard-generator