exa-reference-architecture
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions acknowledge the risks of processing untrusted retrieved content. It mandates defensive measures such as separating retrieved content from operational metadata and explicitly warns against feeding untrusted text into privileged tool execution. The skill includes ingestion points for external data via Exa retrieval products but provides boundary and sanitization guidelines.
- [DATA_EXPOSURE]: The skill provides strong security instructions for authentication, requiring the use of server-side secret managers for
EXA_API_KEY, forbidding the placement of credentials in URLs or code commits, and mandating the use of theAuthorization: Bearerheader. - [SAFE]: The external references in
references/official-docs.mdpoint to the official documentation of Exa.ai, which is a well-known service in the AI search and retrieval space. - [NO_CODE]: The skill contains architectural and instructional markdown only, with no accompanying scripts or executable code provided in the repository.
Audit Metadata