exa-agent

Warn

Audited by Gen Agent Trust Hub on Jul 24, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The 'Batch Script Mode' section instructs the agent to generate and execute local scripts (e.g., for data enrichment) to handle high-volume tasks instead of using repeated tool calls. This involves runtime script creation and execution on the host environment.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of processing external data.
  • Ingestion points: The agent retrieves data from the general web and third-party providers such as Similarweb, Financial Datasets, and Particle News (SKILL.md).
  • Boundary markers: The skill utilizes outputSchema to define structured data requirements, which provides some structural boundary but does not include explicit safeguards against embedded instructions in the retrieved data.
  • Capability inventory: The agent has the ability to write and run local scripts ('Batch Script Mode') and perform recursive web research through Exa Agent runs (SKILL.md).
  • Sanitization: There is no mention of sanitizing, filtering, or escaping instructions found within the ingested web content before it is used for synthesis or script generation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 24, 2026, 05:07 PM
Security Audit — agent-trust-hub — exa-agent