The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided URLs to generate visual screenshots, creating an attack surface for vision-based prompt injection.
Ingestion points: The url parameter in SKILL.md allows the agent to ingest content from arbitrary external websites.
Boundary markers: The skill includes a "Guardrails" section that explicitly instructs the agent to decode the image to a file rather than pasting the base64 data directly into a prompt, which helps mitigate direct injection.
Capability inventory: The skill performs network requests (POST) to api.scavio.dev and provides example code for writing the resulting data to a local file (example.png).
Sanitization: No explicit sanitization of the remote website content is performed by the skill itself, though the conversion to a PNG format provides a structural boundary between the source content and the agent's text context.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill performs network operations to a non-whitelisted external domain.
Evidence: Network requests to api.scavio.dev are documented in the Endpoints and Examples sections of SKILL.md.

website-screenshot-api