The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes content from untrusted external URLs, which creates a surface for indirect prompt injection. This is the primary intended function of the tool. Output is labeled with its source and truncated to prevent excessive resource consumption.
Ingestion points: visit.py (fetches content from external URLs directly and via the Jina Reader service)
Boundary markers: The script prefixes the extracted content with a '## Content from [URL]' header in the output.
Capability inventory: Network read access (HTTP/HTTPS), temporary file writing for images, and printing content to stdout.
Sanitization: The script truncates content over 100KB and performs basic whitespace cleanup.
[DATA_EXFILTRATION]: The script utilizes an optional JINA_API_KEY for authentication with the Jina Reader service. The code ensures this credential is only included in requests to the r.jina.ai domain and is not leaked to the user-provided target URLs during content-type checks or image downloads.

visit-webpage