flexport-core-workflow-a

Pass

Audited by Gen Agent Trust Hub on Sep 13, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data such as origin, destination, and commodity details which could potentially be used for prompt injection.
  • Ingestion points: SKILL.md (Step 1) processes cargo and commodity inputs.
  • Boundary markers: The skill explicitly requires a human-approved booking decision (Step 5) and normalization of inputs (Step 1).
  • Capability inventory: Uses rates_instant_book for mutations, which is strictly gated by the human approval step.
  • Sanitization: Step 1 involves resolving and normalizing inputs through read-only tools before search operations occur.
  • [SAFE]: All external references and network operations target official Flexport domains (flexport.com), which is a well-known service. The author's usage of these resources represents standard integration logic.
  • [SAFE]: The skill enforces secure handling of OAuth tokens and API keys, explicitly forbidding the logging of sensitive credentials and mandating redacted output receipts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 13, 2026, 08:37 AM
Security Audit — agent-trust-hub — flexport-core-workflow-a