ai-core/adapter-configuration
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents patterns for ingesting and processing user-supplied chat data that create an indirect prompt injection surface.\n
- Ingestion points: In SKILL.md, the handleChat and POST examples directly ingest messages from request objects into the model execution context.\n
- Boundary markers: The code snippets lack implementation of delimiters (such as XML tags or special tokens) to isolate untrusted user content from system instructions.\n
- Capability inventory: The skill leverages the chat() and generate() functions to process these messages, enabling the agent to act on potentially malicious instructions embedded in the chat history.\n
- Sanitization: No sanitization, validation, or filtering of the message content is demonstrated in the examples provided for adapter setup.
Audit Metadata