dsh-go-plugin

Warn

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill provides a checker tool in kit/lib/check.js (invoked via kit/bin/check.mjs --load) that uses the import() function to dynamically load and execute Node.js modules from paths authored by the AI. This mechanism executes AI-generated code in the user's local environment without a secure sandbox.
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves the AI generating functional Node.js code which is then executed by the tool's own validation suite, creating a vulnerability surface where malicious instructions could be embedded in generated code.
  • Ingestion points: The lib/index.js file generated by the AI within the plugin directory.
  • Boundary markers: Absent. The code is imported directly as a module with no delimiters or warnings to the executing process.
  • Capability inventory: The validation script runs in the host Node.js environment. While it attempts to mock the dsh-go bridge and sets a temporary DSH_HOME environment variable, it does not use a robust sandbox (like the vm module or a container) to prevent the generated code from performing filesystem, network, or process-level operations.
  • Sanitization: No static analysis or sanitization is performed on the generated Node.js code before it is dynamically imported.
  • [COMMAND_EXECUTION]: The instructions in SKILL.md direct the user to execute shell commands (e.g., mkdir, cp, node) that rely on environment variables and operate on AI-controlled directory paths, which could be exploited if the paths are manipulated.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 19, 2026, 01:05 PM