verify-design-mastery

Warn

Audited by Gen Agent Trust Hub on Oct 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The helper script control-design-mastery is designed to execute repository-local scripts including setup.sh, scripts/check.sh, and scripts/sync-grok-manifest.py. These scripts are executed within the environment of the current repository tree or a temporary clone of a pull request.
  • [EXTERNAL_DOWNLOADS]: When the --pr flag is used, the skill utilizes git to clone the repository and fetch a specific pull request head for verification. Additionally, it invokes the claude and grok command-line interfaces to download and install plugins from the repository or marketplace.
  • [REMOTE_CODE_EXECUTION]: The skill implements a workflow that involves downloading external code (via git clone of a pull request) and subsequently executing scripts (setup.sh, check.sh) contained within that code. While this is the intended purpose of a verification tool, it represents a mechanism for executing untrusted code if the pull request contains malicious content.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes metadata and configuration files (marketplace.json) to determine which plugins to install. The names and values extracted from these files are interpolated into shell commands and used to generate log filenames. This creates an indirect injection surface where a malicious manifest file (e.g., in a compromised pull request) could attempt to influence script execution or perform unauthorized file system operations.
  • Ingestion points: The skill reads plugin names and counts from .claude-plugin/marketplace.json and .grok-plugin/marketplace.json via jq in bin/control-design-mastery.
  • Boundary markers: None present; the skill treats the manifest content as a source of truth for plugin identification.
  • Capability inventory: The skill can write files, create directories, manage symbolic links, perform network operations via git, and execute local shell and Python scripts.
  • Sanitization: The skill performs basic validation by checking requested plugin names against the manifest using grep -qx and utilizes array-based argument passing in its rec function to mitigate standard command injection.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 17, 2026, 09:45 PM