handoff
Pass
Audited by Gen Agent Trust Hub on Jul 19, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill establishes a multi-step indirect prompt injection surface by ingesting conversation history to generate summaries for subsequent agent sessions. Malicious instructions within the source conversation could be summarized and later misinterpreted as legitimate instructions by the next agent.\n
- Ingestion points: The entire current conversation history is ingested to create the summary document.\n
- Boundary markers: The skill does not provide instructions to use clear delimiters or warning tags to help the receiving agent distinguish between summarized context and new instructions.\n
- Capability inventory: The skill utilizes file-write capabilities to save the generated document to the operating system's temporary directory.\n
- Sanitization: Although the skill correctly identifies the need to redact secrets and PII, it lacks measures to identify or filter out instructional overrides or prompt injection payloads.
Audit Metadata