grok-build

Warn

Audited by Gen Agent Trust Hub on Jul 11, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes the grok CLI using the --always-approve flag. This bypasses interactive permission gates, allowing the delegated tool to perform file modifications and run shell commands autonomously on the local system.
  • [REMOTE_CODE_EXECUTION]: By delegating complex implementation tasks to an external agentic CLI with autonomous execution privileges, the skill enables a form of proxy code execution. The delegated agent can perform arbitrary actions on the host machine based on task specifications.
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions to run grok update, which retrieves and installs software updates from xAI's official distribution channels.
  • [PROMPT_INJECTION]: The skill contains an indirect prompt injection attack surface where untrusted repository data could influence the behavior of the delegated agent.
  • Ingestion points: The agent reads and processes data from the target repository specified in the path (SKILL.md).
  • Boundary markers: While instructions are isolated in a temporary task specification file, no markers are used to delimit untrusted repository content from the agent's instructions.
  • Capability inventory: The skill utilizes subprocess calls to grok, git, and filesystem management tools (mkdir, New-Item) across SKILL.md and references/cli.md.
  • Sanitization: No explicit sanitization, validation, or filtering of the repository data is performed before it is processed by the delegated agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 11, 2026, 06:10 PM
Security Audit — agent-trust-hub — grok-build