Loading ToolBench
Local terminal MCP server - gives Claude controlled access to your Windows shell
Four tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. run_command has a dry_run parameter but lacks guidance on AMBER tier behavior. read_file and list_directory have minimal descriptions (under 100 chars). run_approved_script is conditionally loaded and lacks visible schema. Security filtering is mentioned but not formally documented as tool behavior. No error recovery guidance or output field documentation visible.
List directory contents with security filtering
Read file contents with security filtering to prevent access to sensitive files
Execute pre-approved scripts from the personal/approvedScripts module (personal-only feature, conditionally loaded)
Execute a shell command with security filtering (RED tier hard-blocks, AMBER tier warnings, GREEN tier allowed)
run_approved_script has no visible input schema in source code. Tool is conditionally loaded from personal/approvedScripts.js, making schema validation impossible.
read_file and list_directory descriptions are under 100 characters and lack context on when to use them vs alternatives, or what security filtering prevents.
No output schema documented for any tool. LLMs cannot infer what fields are returned (e.g., does read_file return {content, size, encoding}? does list_directory return {files, dirs, total}?).
run_command's dry_run parameter description does not explain AMBER tier behavior or when dry_run is recommended. LLMs cannot determine if they should default to dry_run=true for safety.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
No error recovery guidance. RED tier blocks return structured errors, but descriptions do not tell LLMs what to do next (e.g., 'Command blocked: file deletion prohibited. Use read_file to inspect instead.').