revenue-recognition-accounting
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external datasets such as
subscribers.csvandtransactions.csvto perform accounting calculations. While the provided samples are benign, this ingestion of untrusted data represents an indirect prompt injection surface. Ingestion points:inputs/subscribers.csvandinputs/transactions.csvwithin the evaluation tasks. Boundary markers: Absent. Capability inventory: The agent is tasked with generating and executing Python scripts (recognition_engine.py,sample_calculations.py) based on the input data. Sanitization: No specific sanitization or validation logic is prescribed for the input CSV data. - [DYNAMIC_EXECUTION]: The evaluation tasks require the agent to generate Python scripts and execute them locally to produce accounting reports and journal entries. These scripts perform standard mathematical operations using the
decimalmodule for monetary precision and do not involve untrusted remote sources or library injection.
Audit Metadata