The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The implementation in scripts/main.py explicitly disables global SSL certificate verification by setting ssl._create_default_https_context = ssl._create_unverified_context and calling urllib3.disable_warnings(). While intended as a workaround for certain proxy environments, this removes the primary cryptographic guarantee of the connection's authenticity. An attacker controlling the proxy or network path could intercept the GEMINI_API_KEY and any image data transmitted to Google's servers via a Man-in-the-Middle (MITM) attack.- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted image and prompt data with downstream access to network capabilities.
Ingestion points: Raw image bytes (image_base64) and instructional text (prompt) are ingested via the execute method in scripts/main.py.
Boundary markers: Absent. The skill appends instructions to the prompt but does not use delimiters to isolate user-controlled data from system instructions.
Capability inventory: The skill performs network operations to gemini.googleapis.com using the provided API key.
Sanitization: There is no validation, filtering, or escaping of the image content or prompt input before it is sent to the model.- [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill requires the google-generativeai, pydantic, and pillow libraries. These are official and well-known packages from trusted repositories and do not represent a security risk when obtained from official registries as described in the README.md.

gemini-vision