review-salience-xlsx
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests raw review text from external files or APIs. It lacks explicit boundary markers to isolate review content from agent instructions, creating a surface for indirect prompt injection.
- Ingestion points: load_reviews in scripts/scoring_io.py
- Boundary markers: Absent
- Capability inventory: Network requests (requests), File system writes (openpyxl, json)
- Sanitization: Numeric range and type validation in scripts/scoring_io.py
- [EXTERNAL_DOWNLOADS]: The skill provides documentation and code for integrating with external scoring services like n8n via POST requests to user-defined URLs.
- [DYNAMIC_EXECUTION]: The analysis workflow dynamically modifies the Python path to import its internal processing logic from the scripts directory.
- [DATA_EXFILTRATION]: The skill is designed to send review content to external scoring APIs, which involves transmitting potentially sensitive text to external servers.
Audit Metadata