quiet-shell

Warn

Audited by Gen Agent Trust Hub on Feb 27, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads the @codemcp/quiet-shell package from the npm registry during execution.
  • [REMOTE_CODE_EXECUTION]: The package is executed using npx -y, which bypasses user confirmation and runs code from a package scope (@codemcp) that is not affiliated with the skill author (mrsimpson) or recognized as a trusted organization.
  • [COMMAND_EXECUTION]: Because the tool is designed to wrap shell commands, the unverified remote code has direct interaction with the host system's shell environment.
  • [PROMPT_INJECTION]: The skill's primary function of processing and suppressing shell output creates a vulnerability to indirect prompt injection.
  • Ingestion points: Standard output and error streams from shell commands executed via the skill in SKILL.md.
  • Boundary markers: No specific delimiters or instructions exist to notify the agent to ignore instructions that might be embedded in the filtered command output.
  • Capability inventory: The skill is explicitly granted the capability to execute shell commands using the quiet-shell tool.
  • Sanitization: There is no evidence of output validation or sanitization to prevent the agent from interpreting command output as instructions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 27, 2026, 07:33 AM