oneroster-csv-validator
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted OneRoster CSV data (such as student names, course titles, and organizational data) which could contain malicious prompts designed to manipulate the agent's logic during validation.
- Ingestion points: The skill ingests data from several CSV files (
manifest.csv,orgs.csv,users.csv,courses.csv,classes.csv,enrollments.csv, andacademicSessions.csv) typically provided in a ZIP archive. - Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" markers when the agent processes or reports on the content of these CSV fields.
- Capability inventory: The skill provides a Python implementation that uses
zipfile,csv, andiomodules to extract and parse files. It does not utilize network operations or arbitrary command execution tools. - Sanitization: The provided Python logic includes robust sanitization and security checks to prevent Denial of Service (DoS), such as limiting the number of files in the archive (100), total expansion size (16MB), individual file size (4MB), compression ratios (100:1), and maximum CSV row counts (10,000). It also enforces strict UTF-8 decoding and strips Byte Order Marks (BOM).
Audit Metadata