context-retrieval
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a Retrieval-Augmented Generation (RAG) workflow that ingests untrusted data from external sources into the agent's prompt sequence.
- Ingestion points: Data is retrieved from external vector databases and keyword indices as described in the 'Workflow' section of SKILL.md.
- Boundary markers: The skill recommends using system instructions to ground the model ('answer only from the provided context') to mitigate hallucinations and instruction override, though it lacks physical delimiters for retrieved chunks.
- Capability inventory: No direct system capabilities (file writes, network, shell) are defined in the provided file, but the skill is designed to influence the agent's generation phase.
- Sanitization: The workflow does not specify sanitization, escaping, or filtering of the retrieved document content before prompt interpolation.
Audit Metadata