The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for file exfiltration due to unvalidated file path inputs.\n
Ingestion points: The --ref argument in scripts/generate.py accepts arbitrary file paths from the agent's context, which can be influenced by user prompts or data.\n
Boundary markers: The skill lacks technical boundary markers or instructions to isolate potentially malicious embedded commands in processed inputs.\n
Capability inventory: The scripts/generate.py script possesses the capability to read local files using open() and transmit their contents (base64-encoded) to the OpenRouter API via urllib.request.\n
Sanitization: The script does not perform path validation or file type verification on the reference images, allowing it to read and transmit any file accessible to the agent process if provided as a reference path.\n- [EXTERNAL_DOWNLOADS]: The skill interacts with the well-known service OpenRouter (openrouter.ai) to generate and retrieve images.\n
The fetch function in scripts/generate.py retrieves image data from URLs provided by the API response. This is standard functionality for image generation tools.\n- [SAFE]: API key management follows standard security practices.\n
The skill resolves keys from environment variables or a gitignored key.txt file, minimizing the risk of accidental credential exposure in version control.

image-gen