inflection

Pass

Audited by Gen Agent Trust Hub on Aug 2, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill ingests untrusted data from SEC filings, creating an indirect prompt injection surface.
  • Ingestion points: Management commentary is retrieved and summarized from external SEC filings in step 6.
  • Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when handling this external data.
  • Capability inventory: The agent has the capability to write local files and call discovery tools.
  • Sanitization: No sanitization or escaping of the commentary is mentioned before it is embedded into the final HTML report output.- [COMMAND_EXECUTION]: The skill constructs local file system paths using unvalidated user input.
  • Evidence: In step 8, the skill saves a report to a path derived directly from the 'TICKER' argument ($ARGUMENTS) without explicit validation or sanitization of the input string.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 2, 2026, 07:47 PM
Security Audit — agent-trust-hub — inflection