The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because it is designed to ingest and process content from arbitrary public web sources.
Ingestion points: External web data is brought into the agent's context through the exa_search and exa_contents tools as described in SKILL.md and references/sandbase-api-map.md.
Boundary markers: The instructions do not define clear delimiters or include specific guidance for the agent to ignore instructions or commands embedded within the retrieved search results or extracted text.
Capability inventory: The skill uses the sandbase_call_tool to execute search and extraction operations.
Sanitization: The skill does not implement or mention sanitization, filtering, or validation of the content retrieved from external sources before presenting it to the agent for synthesis.

exa-deep-search