manage-order-fulfillment
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process webhook callbacks from external logistics providers (GHN, GHTK, GrabExpress, etc.), which represents an indirect prompt injection surface.
- Ingestion points: Tracking updates and COD status webhooks (SKILL.md, UPGRADE_PLAN_2026.md).
- Boundary markers: No explicit instructional boundaries or delimiters are specified for the incoming data.
- Capability inventory: The skill utilizes
write_file,edit_file, andfetchtools. - Sanitization: The instructions mandate HMAC signature verification and the use of a carrier allowlist to validate incoming webhook integrity.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill uses the
fetchtool to communicate with third-party carrier APIs and receive status updates. These network operations target domains specific to Vietnamese logistics services that are not on the standard whitelist. However, the skill includes explicit rules to ensure customer PII (address, phone) is treated as confidential and excluded from logs.
Audit Metadata