field-service-data-capture-form-designer-configure

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, specifically user-provided PDF files, images, and prose descriptions, to generate functional Salesforce metadata.
  • Ingestion points: The skill reads file paths provided by the user for .pdf, .png, .jpg, and .jpeg files using the Read tool, as well as natural language descriptions.
  • Capability inventory: The agent can write JSON specifications to /tmp/data-capture-spec.json and invoke a secondary deployment skill that performs Salesforce REST API operations (POST /services/data/vXX.0/tooling/sobjects/Flow) to create or update metadata in the user's environment.
  • Sanitization: The skill enforces Salesforce API naming conventions (PascalCase/camelCase) for generated fields but does not explicitly mention filtering instructions embedded within the processed document text.
  • Boundary markers: The skill relies on a mandatory human-in-the-loop confirmation step (Step 4) rather than technical delimiters. The agent is strictly instructed to use AskUserQuestion to present the full plan, including object mappings and field types, to the user before proceeding to the deployment phase.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 06:38 PM
Security Audit — agent-trust-hub — field-service-data-capture-form-designer-configure