The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input via the --prompt argument which is directly interpolated into the prompt sent to the Gemini API.\n
Ingestion points: The --prompt argument in SKILL.md and scripts/image.py (line 155).\n
Boundary markers: The script prefixes the user prompt with internal instructions regarding aspect ratios (lines 101-110), but lacks strict delimiters to prevent the user input from overriding these instructions.\n
Capability inventory: The skill can create directories (os.makedirs line 144), write generated images to disk (open line 153), and read local image files (Image.open line 117).\n
Sanitization: There is no validation or sanitization of the prompt string before it is sent to the model API.\n- [COMMAND_EXECUTION]: The skill instructions involve executing shell commands to manage environment variables and run the generation script.\n
The instructions provide a specific grep command to load the GEMINI_API_KEY from a .env file in the project root. This is a targeted access to a sensitive configuration file, though the instructions include safety warnings to avoid accidental leakage of other environment variables.

generate