email-invoice-processor

Pass

Audited by Gen Agent Trust Hub on May 13, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses sensitive user data by connecting to email accounts via IMAP and reading message contents. It also stores and reads email credentials (authorization codes) from a local configuration file (config.json) to perform its primary function.
  • [EXTERNAL_DOWNLOADS]: The process_invoices.py script automatically downloads files from URLs extracted from email bodies. It interacts with external endpoints using both the requests library and a headless browser (playwright) to retrieve invoice documents.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data from emails and external websites.
  • Ingestion points: Email headers, email bodies, and external web page content processed by scripts/process_invoices.py.
  • Boundary markers: Absent; untrusted data is processed without specific delimiters or instructions to ignore embedded commands.
  • Capability inventory: File system write access for saving invoices, Excel summaries, and logs; network access for IMAP connectivity and HTTP downloads.
  • Sanitization: The script performs magic bytes verification (e.g., checking for %PDF or image signatures) to ensure downloaded content matches expected file types before storage. It uses regular expressions and table extraction for data parsing.
Audit Metadata
Risk Level
SAFE
Analyzed
May 13, 2026, 01:04 PM
Security Audit — agent-trust-hub — email-invoice-processor