handoff-fusion-engineering
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [Indirect Prompt Injection Surface]: The skill processes external engineering data and evidence metadata, which can be an entry point for untrusted instructions. \n
- Ingestion points: Data is ingested from external evidence references and via automated inspection tools like
document.inspectandcam.inspect(SKILL.md). \n - Boundary markers: The instructions do not define explicit delimiters to separate user-provided engineering data from the agent's core instructions. \n
- Capability inventory: The skill uses tools to prepare drafts (
fusion_handoff_prepare) and perform geometric or CAM inspections (geometry.check,parameters.list) (SKILL.md). \n - Sanitization: No specific sanitization or filtering mechanisms for the ingested data are described, although the skill notes that external evidence is unfetched and unverified. \n- [Sensitive Data Protection]: The skill instructions explicitly warn against selecting credential fields or relying on redaction placeholders for data comparison. This helps mitigate the risk of accidental exposure of authentication secrets during the engineering review process. \n- [Operational Isolation]: The skill is designed for local draft preparation only and lacks capabilities to publish artifacts, send drafts, or update enterprise systems, which acts as a safeguard against unauthorized external data transmission.
Audit Metadata