cohere-core-workflow-a
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a complete Retrieval-Augmented Generation (RAG) pipeline, which serves as a common vector for indirect prompt injection attacks.
- Ingestion points: The workflow ingests external document corpora and user queries as primary inputs for embedding, reranking, and generation steps (SKILL.md).
- Boundary markers: The implementation utilizes Cohere's structured
documentsparameter in the Chat API, which provides a structural delimiter between retrieved context and the conversation history. - Capability inventory: The skill is configured with tools for file reading, writing, and bash execution (for npm packages), alongside the network capabilities required for the Cohere API.
- Sanitization: The provided code snippets do not include content sanitization or verification steps for the knowledge base documents before they are processed by the LLM.
Audit Metadata