The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes content from untrusted external URLs, creating a surface for indirect prompt injection. Malicious web pages could contain hidden or explicit instructions designed to subvert the agent's logic when the content is eventually read by the language model.
Ingestion points: scripts/read.py takes a URL from the input_data dictionary and retrieves content via the WebIntelligenceRouter.
Boundary markers: No specific delimiters (such as XML tags or boundary strings) are added to the output to isolate external content from system instructions.
Capability inventory: The skill utilizes kernel.web for network operations and data extraction, returning the result as a structured JSON object.
Sanitization: No sanitization, HTML stripping, or safety filtering is performed on the content within the skill's logic before it is returned.

web-read