driver-based-budget-builder
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted data via the
--configJSON input inscripts/driver_budget.py. While the script currently performs numerical calculations, the lack of robust sanitization or strict schema enforcement at the ingestion point could allow a malicious actor to supply malformed data. If the output of this script is subsequently fed into another agent or used to influence further automated decisions (as suggested in Step 7 ofSKILL.md), it could lead to multi-step injection chains or bias the agent's financial analysis. - Ingestion points:
scripts/driver_budget.pyreads a JSON file specified by the--configargument. - Boundary markers: None. The script directly loads and processes the JSON content.
- Capability inventory: The script performs file reading (
open), JSON parsing (json.load), and standard output printing. - Sanitization: None. The script assumes the JSON follows the expected structure and contains valid numerical values for growth rates and costs.
Audit Metadata