hedgefund-harvard-dividend
Warn
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: MEDIUMCOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using hardcoded absolute file paths (e.g.,
/Users/engineer/.venv/bin/python3). This pattern exposes internal system directory structures and assumes a specific local environment configuration, which can be a security risk in shared or multi-user environments. - [COMMAND_EXECUTION]: The instructions direct the agent to execute Python code dynamically via the
python3 -cflag. This execution model is vulnerable to command injection if parameters such as stock tickers or external data are not correctly sanitized before being concatenated into the shell command. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources (
yfinance,web_fetch) and user-provided inputs. 1. Ingestion points: external financial data and user goal parameters in SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: Shell command execution and Python script execution in SKILL.md. 4. Sanitization: Absent. It does not implement technical delimiters or data sanitization to protect against malicious instructions embedded in the external content.
Audit Metadata