datasets

Pass

Audited by Gen Agent Trust Hub on Nov 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill manages and processes data from various untrusted external sources, including local files (CSV, JSON, Excel) and live databases. This data is subsequently utilized in test runs and made available to scripts, representing a surface for indirect prompt injection if the source data contains adversarial content.
  • Ingestion points: External data enters the agent context via manifests defined in .dataset.yaml files and through external sources attached using the postman dataset source add command (SKILL.md, reference/schema.md).
  • Boundary markers: The instructions advocate for "credential boundaries" and the use of "shared-vault references" (SKILL.md), which helps isolate secrets, though no specific markers are mentioned for separating the data content from instructions.
  • Capability inventory: The skill facilitates SQL query execution via postman dataset query and allows the pm.datasets() interface to expose external data to Javascript snippets during collection runs (SKILL.md).
  • Sanitization: There is no evidence in the instructions or schema reference that external data is sanitized or validated against a schema before being processed by the agent or the script environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 3, 2026, 08:29 PM