The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user-provided prompts and images which are then processed by an LLM-based image generation model. It lacks explicit boundary markers or instructions to isolate potentially malicious commands embedded within the input data. • Ingestion points: The scripts/generate_image.py script accepts user-supplied data via the --prompt and --input-image arguments. • Boundary markers: No specific delimiters or instructions to ignore embedded instructions are present in the skill instructions. • Capability inventory: The skill is granted terminal execution permissions (run_terminal) and filesystem write access (write_file and Python Path operations). • Sanitization: The script processes input parameters directly and uses pathlib for filename handling without validating against path traversal attacks.
[COMMAND_EXECUTION]: The skill uses the uv run command to execute a local Python script with arguments constructed from user input. This pattern creates a potential execution path where an agent could be manipulated into running commands with malicious parameters, such as escaped characters or sensitive path targets.
[EXTERNAL_DOWNLOADS]: The skill utilizes the google-genai library to interact with Google's official Gemini API services. This is a legitimate requirement for the skill's functionality and involves network communication with a well-known service provider.

inno-figure-gen