The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided natural language prompts to determine image cropping parameters, creating a potential surface for indirect prompt injection.
Ingestion points: The prompt parameter in photo_clipper.py and the main entry function.
Boundary markers: The skill lacks explicit delimiters or specific instructions to the LLM to ignore potentially malicious commands within the user's prompt during vision analysis.
Capability inventory: The skill can read local image files, write new cropped image files to the file system, and perform network requests to an external API.
Sanitization: It implements verification checks to ensure files are JPEG/PNG and utilizes Pydantic fields to limit the cropping area to a maximum of 50% of the original dimensions.
[EXTERNAL_DOWNLOADS]: The skill communicates with the OpenRouter API to process vision analysis requests.
Evidence: The implementation uses the OpenAI client to connect to the well-known https://openrouter.ai/api/v1 endpoint as documented in the source code.

photo-clipper