autoresearch
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill requires the agent to run user-defined shell commands for evaluation (e.g., 'uv run backtest.py'). This involves executing arbitrary code in the environment.
- [DYNAMIC_EXECUTION]: The core loop involves the agent proposing and implementing code modifications, which are then immediately executed. This 'write-then-run' pattern is a form of dynamic execution.
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface.
- Ingestion points: Data enters the loop via the 'target file' (modified by the agent) and the 'eval command' output.
- Boundary markers: None identified; there are no delimiters separating the untrusted code output from the agent's reasoning.
- Capability inventory: The skill uses Bash, Write, and Edit tools to modify files and execute commands.
- Sanitization: No validation or sanitization of the generated code or evaluation output is performed before re-ingestion.
- [PROMPT_INJECTION]: The instructions explicitly direct the agent to bypass standard interactive safety checks ('CONTINUE — do NOT stop to ask the user') and maintain execution indefinitely ('NEVER: ... Stop the loop without being interrupted').
Audit Metadata