setup
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes multiple user-provided inputs including 'eval_cmd', 'domain', and 'name' which are interpolated into a shell command. If these inputs are sourced from untrusted data, it could lead to unauthorized command execution.\n
- Ingestion points: User-provided arguments for the '/ar:setup' command and interactive prompts in 'SKILL.md'.\n
- Boundary markers: None present in the instructions to distinguish between trusted and untrusted input.\n
- Capability inventory: Executes a Python script ('setup_experiment.py') which runs an arbitrary evaluation command provided by the user.\n
- Sanitization: No sanitization or validation of the input strings is mentioned.\n- [COMMAND_EXECUTION]: The skill constructs and executes a shell command ('python setup_experiment.py ... --eval "{eval_cmd}"') using a string provided by the user. This is deliberate functionality but allows for the execution of arbitrary code within the shell context.\n- [DYNAMIC_EXECUTION]: The command string is dynamically assembled at runtime based on interactive user input or command-line arguments, which is a common pattern for injection vulnerabilities if the resulting string is executed by a shell without proper escaping.
Audit Metadata