zj-handoff
Pass
Audited by Gen Agent Trust Hub on Jul 27, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect prompt injection surface identified. The skill processes potentially untrusted conversation history and user arguments to create a summary for a subsequent agent. This creates a chain where malicious instructions could be passed between sessions.
- Ingestion points: Conversation history and user-supplied arguments.
- Boundary markers: The instructions do not specify delimiters or warnings to separate summarized untrusted content from agent instructions.
- Capability inventory: File system write operations to the operating system's temporary directory.
- Sanitization: Instructions include a requirement to redact sensitive credentials (API keys, passwords) and PII, which mitigates data exposure but not logic-based injection.
Audit Metadata