grok
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill documents an installation procedure that fetches a setup script from xAI's official domain via
curl -fsSL https://x.ai/cli/install.sh | bash. - [EXTERNAL_DOWNLOADS]: The skill references downloading the Grok CLI through the official
@xai-official/grokpackage on the NPM registry and thex.aidomain. - [COMMAND_EXECUTION]: The skill instructions utilize the
terminal()tool to invoke thegrokCLI, along with standard development tools likegit,tmux, andghfor managing repositories and pull requests. - [INDIRECT_PROMPT_INJECTION]: The skill demonstrates a surface for indirect prompt injection where user-supplied tasks are interpolated into the
grokcommand line. - Ingestion points: User-provided strings are passed as arguments to the
grok -pcommand as shown in the examples withinSKILL.md. - Boundary markers: Commands use single quotes to delimit the prompt (e.g.,
grok -p 'Add a dark mode toggle'). - Capability inventory: The skill uses the
terminal()tool to executegrok, an autonomous coding agent with full file system access and shell execution capabilities. - Sanitization: The documented examples do not include explicit input validation or sanitization for the prompts passed to the CLI.
Audit Metadata