The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external text prompts that are sent to the OpenRouter API, creating a surface for indirect instructions to influence agent behavior if the prompt source is untrusted.\n
Ingestion points: The prompt positional argument in scripts/generate_image.py and the prompt instructions in SKILL.md.\n
Boundary markers: None present; the prompt is interpolated directly into the API request payload.\n
Capability inventory: The skill has network capabilities (POST requests to OpenRouter) and file system access (reading input images, writing output PNGs).\n
Sanitization: No content filtering or validation is implemented on the prompt string.\n- [COMMAND_EXECUTION]: The skill instructions direct the agent to execute a local Python script (scripts/generate_image.py) with user-provided arguments.\n- [DATA_EXFILTRATION]: The script transmits text prompts and base64-encoded image data to the OpenRouter API endpoint at https://openrouter.ai/api/v1/chat/completions. This is the intended behavior for image processing via a well-known service.\n- [CREDENTIALS_UNSAFE]: The script scripts/generate_image.py and the instructions in SKILL.md facilitate the retrieval of the OPENROUTER_API_KEY from .env files located in the current or parent directories. This follows standard development practices for secret management.

generate-image