grok-coding-agent
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes the
grokCLI tool to perform coding tasks. This is the primary function of the skill and is handled using secure defaults. - It explicitly mandates the use of
--sandboxprofiles to restrict file system access. - It utilizes
--permission-mode dontAskin conjunction with narrow--allowand--denyflags to ensure the agent cannot perform actions outside the predefined scope. - It encourages the use of read-only sandboxes for information-gathering tasks to prevent accidental modifications.
- [PRIVILEGE_ESCALATION]: The skill instructions specifically warn against trusting unfamiliar repository profiles and require verification of host isolation. It notes that macOS read-only modes do not inherently block network access, advising the user to preserve network boundaries separately.
- [DATA_EXFILTRATION]: No patterns of automated data exfiltration were detected. The skill focuses on local execution within user-defined workspaces and provides guidance on filtering tools to prevent unauthorized data access via MCP or subagents.
Audit Metadata