hedgefund-bridgewater-risk
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes shell commands in SKILL.md to perform risk calculations. Specifically, it executes
bun .agents/skills/risk-desk/scripts/risk-desk.tsand/Users/engineer/.venv/bin/python3 .agents/skills/stocks-advisor/scripts/fundamentals.py. These commands utilize arguments based on user-supplied portfolio information, which could be exploited for command injection if input tickers or file names are not strictly validated.\n- [PROMPT_INJECTION]: The skill processes external information through theweb_fetchtool and user input, creating a risk for indirect prompt injection in SKILL.md.\n - Ingestion points: Data retrieved from external URLs via
web_fetchand portfolio details provided by the user.\n - Boundary markers: None. The instructions lack delimiters to segregate external data from logic, increasing the risk of the agent following instructions embedded in the data.\n
- Capability inventory: Access to shell execution (
bun,python3) and file system operations (writing CSV files).\n - Sanitization: No sanitization or filtering logic is specified for data fetched from the web or provided by the user before it affects script execution.
Audit Metadata