poffice-admin

Pass

Audited by Gen Agent Trust Hub on Nov 1, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources which could contain malicious instructions intended to influence the agent's behavior.
  • Ingestion points: The skill reads email headers and subjects in scripts/poffice_imap.py and retrieves user records from a Neon PostgreSQL database in scripts/poffice_master_admin.py.
  • Boundary markers: None identified. Instructions are not delimited from data.
  • Capability inventory: The skill has broad capabilities including executing subprocesses (scripts/poffice_master_admin.py), writing files (scripts/poffice_gen.py), and performing network operations across all core scripts.
  • Sanitization: There is no evidence of sanitization or filtering for the data retrieved from IMAP or the database before it is processed or used in logic.
  • [COMMAND_EXECUTION]: The script scripts/poffice_master_admin.py uses subprocess.check_output to execute other Python scripts within the skill.
  • Evidence: cmd = ["python3", "-W", "ignore", ".agents/skills/poffice-admin/scripts/poffice_mail.py", ...]
  • Context: While it uses a list format (reducing shell injection risk), the arguments like email, name, and password are sourced from external database records, creating a path for command argument manipulation if the called scripts do not handle inputs securely.
  • [DATA_EXFILTRATION]: The skill handles sensitive credentials including the Mailcow API key, Seafile admin passwords, and Paperless-ngx tokens.
  • Risk: The skill disables SSL/TLS certificate verification (verify=False) in scripts/poffice_mail.py, scripts/poffice_docs.py, and scripts/poffice_paperless.py. This exposes sensitive administrative credentials to Man-in-the-Middle (MitM) attacks if the agent operates on an untrusted network.
  • [EXTERNAL_DOWNLOADS]: The skill interacts with external services to manage data.
  • Details: It downloads files from a Seafile instance (cloud.poffice.online) and fetches configuration/user data from a Neon database and Mailcow API (mail.poffice.online). These operations are consistent with the skill's primary purpose.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 1, 2026, 08:41 AM