skills/edwinhu/workflows/crsp-v2/Gen Agent Trust Hub

crsp-v2

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-supplied parameters such as dates, frequencies, and specific query constraints to generate SQL commands executed against an external database.
  • Ingestion points: User-provided arguments for time ranges (start/end dates) and data frequency (daily/monthly) which are interpolated into SQL templates within SKILL.md and examples/ciz_panel.py.
  • Boundary markers: The documentation uses specific XML-like tags (e.g., ) to separate critical data integrity rules from general instructions, though it lacks explicit delimiters for untrusted user input within the generated queries.
  • Capability inventory: The skill utilizes the psycopg2 library to establish network connections to the WRDS PostgreSQL server and execute SQL queries, as shown in examples/ciz_panel.py.
  • Sanitization: The provided Python example follows security best practices by using parameterized queries (e.g., %(start)s) via psycopg2 to prevent SQL injection, rather than direct string concatenation of user input.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 05:55 AM
Security Audit — agent-trust-hub — crsp-v2