hedge-backtesting

Warn

Audited by Gen Agent Trust Hub on Aug 20, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill performs dynamic code generation by translating user-provided strategy rules into a Python script (backtest_<slug>.py) and then executing it via a subprocess. This creates a vector for arbitrary code execution if the input rules are manipulated into containing malicious logic.
  • [COMMAND_EXECUTION]: The agent is instructed to execute shell commands using specific system paths and command-line utilities: /Users/engineer/.venv/bin/python3 .cache/hedge-backtesting/<slug>/backtest_<slug>.py 2>&1 | tee .cache/hedge-backtesting/<slug>/run.log. The use of hardcoded environment paths reveals information about the host system structure.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes external, untrusted strategy rules and interpolates them into executable code.
  • Ingestion points: Strategy rules are ingested from the user or the hedge-strategy-generation skill (SKILL.md).
  • Boundary markers: The instructions lack formal delimiters or "ignore instructions" warnings when processing these rules.
  • Capability inventory: The agent has the capability to write files, execute shell commands, and perform network requests through the yfinance library.
  • Sanitization: There is no specified sanitization or validation logic to ensure the rules do not contain malicious code or shell escapes before being written to the script.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 20, 2026, 02:21 PM
Security Audit — agent-trust-hub — hedge-backtesting