ai-provider-cohere-sdk
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents patterns for Retrieval-Augmented Generation (RAG) and tool use, which involves the ingestion of untrusted external data that could potentially contain malicious instructions.
- Ingestion points: Untrusted data enters the agent context via the
documentsparameter in thechat()method and throughtoolrole messages containing function results, as described inexamples/tools-rag.mdandreference.md. - Boundary markers: The skill demonstrates the use of structured objects (
{ data: { ... } }) which helps the SDK maintain data boundaries, and identifies thesafetyModeparameter in the API reference to help manage content safety. - Capability inventory: The documented patterns include network operations to the official Cohere API and the execution of tool functions.
- Sanitization: The documentation focuses on SDK implementation patterns; it is recommended that developers implement source-level validation and sanitization when ingesting content from untrusted external sources.
Audit Metadata