analyze
Warn
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill facilitates the execution of dynamic financial formulas and 'factors'. It accepts Python callables (PrefixRecomputeFn) and executes them against financial datasets within the AnalyzeFacade. Although it attempts to enforce a trust boundary using 'BoundPrefixRecompute' and time-causality checks, it remains a vector for dynamic code execution. (Files: facade.py, causality.py).
- [INDIRECT_PROMPT_INJECTION]: The report generation module in tearsheet.py constructs HTML summaries by interpolating untrusted factor metadata. Specifically, the factor_id field is embedded directly into HTML figcaption elements without sanitization, creating a surface for HTML injection or cross-site scripting (XSS) when reports are rendered.
- Ingestion points: Data entries from factor_test_summary parquet files (File: tearsheet.py).
- Boundary markers: Absent in the HTML generation f-strings.
- Capability inventory: File-system write access via Path.write_text and matplotlib.savefig (File: tearsheet.py, ts_analysis.py).
- Sanitization: Filenames are sanitized, but metadata embedded in the HTML body is not.
Audit Metadata