The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted search result data from the Exa neural search API, creating a surface for indirect prompt injection attacks where malicious instructions in indexed web content could influence agent behavior.
Ingestion points: Data is ingested via exa.searchAndContents and processed into snippets in the processForRAG and ragSearch functions within SKILL.md.
Boundary markers: The ragSearch implementation uses simple numeric markers (e.g., [1]) and horizontal rule separators (---) to delimit results, but does not include explicit instructions for the agent to disregard commands or instructions contained within the external text snippets.
Capability inventory: The skill uses the Read, Write, and Edit tools to manage data and potentially update files.
Sanitization: The skill implements character-based truncation (.slice(0, maxSnippetLength)) but lacks sanitization or filtering to remove potential instruction-carrying markup or text from the search results.

exa-data-handling