exa-advanced-troubleshooting
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines functions that ingest untrusted strings (queries and URLs) which are then processed by the Exa SDK and external network requests, creating a potential vector for indirect instructions.
- Ingestion points: The
queryparameter inprofileLatencyand theurlparameter indebugContentRetrievalfound inSKILL.md. - Boundary markers: Absent. The inputs are passed directly to the SDK methods without delimiters or instructions to ignore embedded commands.
- Capability inventory: The skill performs network operations via
fetchand multipleexa-jslibrary methods (search,searchAndContents,findSimilar,getContents) inSKILL.md. - Sanitization: Absent. The code snippets do not include logic to validate or sanitize the input strings before they are used in network-bound API calls.
Audit Metadata