research-add-fields

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from web searches and user input, which introduces a potential surface for indirect prompt injection.\n
  • Ingestion points: Step 2 ingests data via web searches and direct user input into the agent session.\n
  • Boundary markers: The prompt template lacks distinct boundary markers or instructions to isolate the untrusted external content.\n
  • Capability inventory: The skill reads from and appends data directly to the local fields.yaml file.\n
  • Sanitization: No input validation, escaping, or structural sanitization is performed on the gathered fields before writing them to the configuration file.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 12:39 PM
Security Audit — agent-trust-hub — research-add-fields