The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from the internet, creating a potential surface for indirect prompt injection where malicious instructions embedded in web pages could influence the agent's behavior.
Ingestion points: Untrusted content retrieved from external websites using WebFetch and WebSearch in SKILL.md.
Capability inventory: The skill utilizes Write and Edit for file operations, and WebSearch/WebFetch for network access.
Boundary markers: The instructions emphasize source attribution and factual verification, but do not provide explicit structural delimiters (like XML tags or specific block quotes) to isolate untrusted web content from the agent's internal instruction set.
Sanitization: No specific content sanitization or instruction-filtering mechanisms are described in the research process.

research-expert