language-runtimes
Pass
Audited by Gen Agent Trust Hub on Aug 21, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill provides instructions for installing various development tools using piped remote script execution. Specifically, it references the official installation methods for NVM (via GitHub), SDKMAN (via sdkman.io), and Rustup (via rustup.rs) using the curl | bash or curl | sh pattern. These patterns are used to fetch and execute scripts directly in the shell.\n- [COMMAND_EXECUTION]: The verification script scripts/check.sh performs dynamic evaluation of command strings. It uses the eval function to execute hardcoded version-checking commands such as node -v and python3 --version. While the input strings are currently hardcoded within the script's logic, the use of eval for command execution is a noted security pattern.\n- [PROMPT_INJECTION]: The skill ingests data from external configuration files, creating a surface for indirect prompt injection.\n
- Ingestion points: The skill reads ~/.zshrc, ~/.zprofile, .nvmrc, and .node-version to configure the environment.\n
- Boundary markers: There are no explicit instructions for the agent to ignore or delimit potentially malicious content within these files.\n
- Capability inventory: The skill allows the agent to execute shell commands, perform network operations, and write to shell profile files.\n
- Sanitization: Content from the ingested configuration files is processed without validation or sanitization.
Audit Metadata