personal-finance-hardened
Pass
Audited by Gen Agent Trust Hub on Apr 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill includes a Python script (scripts/init_db.py) designed to initialize the local SQLite database. This script creates the necessary directory structure and tables within a specific workspace path in the user's home directory.
- [DATA_EXFILTRATION]: The skill instructions contain explicit security guardrails that forbid the transmission of sensitive financial data to external services. The architecture is designed for local-only processing to maintain data privacy.
- [PROMPT_INJECTION]: The skill handles user-provided data such as transaction descriptions which are stored in a database and retrieved for summaries, creating a potential surface for indirect prompt injection.
- Ingestion points: User-provided transaction descriptions and category names logged into finance.db.
- Boundary markers: Explicit 'Security Guardrails' section in SKILL.md defining validation rules and operational limits.
- Capability inventory: SQLite database operations (INSERT, SELECT, UPDATE) and local file system access.
- Sanitization: Instructions include validation for positive numeric amounts and restricted SQL operation sets to mitigate abuse.
- [EXTERNAL_DOWNLOADS]: The skill documentation references external URLs for safety reports and feedback on the vendor's official platform and GitHub. These references are for documentation and feedback purposes and do not involve runtime code execution.
Audit Metadata