grok-deep-research
Pass
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes the
grokCLI binary via thesubprocess.runfunction inscripts/grok_deep_research.py. The implementation uses a list-based argument structure and avoids the use ofshell=True, effectively preventing shell injection vulnerabilities at the host level. - [PROMPT_INJECTION]: Research prompts are constructed by interpolating user-provided queries. The skill implements a robust security layer by passing
--denyflags (denyingBash(*),write_file,edit_file, andapply_patch) to thegrokCLI. This ensures the Grok agent remains in a restricted, read-only research mode even if the processed content contains adversarial instructions. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from the web. However, the potential impact is mitigated by the mandatory exclusion of high-privilege tools (shell, file writing) in the
grokCLI invocation, reducing the agent's capability to perform harmful actions based on external content.
Audit Metadata