qonto-fisca-copilot

Pass

Audited by Gen Agent Trust Hub on Oct 9, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill uses read-only tools to access bank transactions and organization details, preventing any unauthorized modification of financial data.
  • [SAFE]: A localized Python script is used to process transaction data deterministically. This approach ensures that sensitive financial categorization is performed by pre-defined code rather than relying solely on AI reasoning, which reduces the risk of manipulation.
  • [SAFE]: The instructions include specific, high-quality defensive guardrails against indirect prompt injection. It explicitly warns the agent that content fetched from receipts or web searches must be treated as untrusted data and never as instructions, with mandatory reporting of any detected command-like text to the user.
  • [SAFE]: Network operations are restricted to the vendor's official AWS S3 bucket for downloading receipt attachments. The skill correctly identifies presigned URLs as sensitive credentials and provides instructions to prevent their exfiltration or logging.
  • [SAFE]: External lookups via web search are opt-in and limited to sanitized public merchant names, requiring explicit user confirmation before any data is sent to external search providers.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 9, 2026, 02:27 PM