Loading ToolBench
An AI-powered code CLI tool that provides agent-based code execution, file manipulation, shell commands, and MCP resource integration
x-code-cli exposes 21 tools with significant definition gaps. Most tools lack complete input schemas visible in source; descriptions are present but often generic (10-50 chars). Parameter types are declared in JSON but descriptions are sparse or missing. No output schemas documented. Error handling is minimal. Tool naming follows verb_noun convention (askUser, readFile, shell) but lacks clarity on mutual exclusivity and dependencies. The server mixes high-risk operations (shell, writeFile, edit) with read-only tools without clear permission boundaries or confirmation patterns.
Ask the user a question with optional multiple choice options
Capture and analyze a browser screenshot for visual QA
Edit a file by replacing old string with new string or applying batch edits
Enter plan mode to create and review a plan before implementation
Exit plan mode and request user approval for the plan
Find files matching a glob pattern
Search for text patterns in files
sendMessage tool has empty input schema (no parameters documented) and minimal description (30 chars). LLM cannot determine what message content or recipient fields are required.
browserVisualCheck and shellOutput lack input schema visibility and descriptions under 50 chars. Cannot infer parameter semantics or when to invoke these tools.
No output schemas documented for any tool. LLMs cannot plan downstream tool calls or extract required fields (e.g., does readFile return {content, size, encoding}?). Violates pattern:tool and mxe:include-chaining-ids.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
List available peer agents from the peer service
List directory contents with optional recursive traversal
List available resources from connected MCP servers, optionally filtered by server name
Read the contents of a file
Read the content of a specific MCP resource by URI
Send a prepared message to a peer agent
Execute shell commands with optional working directory, timeout, and output limits
Passive shell output monitoring tool
Execute a sub-agent task with a specific agent type and description
Create or update a todo checklist with items having content, activeForm, and status fields
Search for and activate tools from a deferred catalog based on a query
Fetch content from a web URL
Search the web for information
Create or overwrite a file with the specified content
Destructive tools (shell, writeFile, edit) lack confirmation patterns, dry-run support, or explicit permission gates. No error recovery guidance. Agents can execute irreversible operations without safeguards.
Parameter descriptions are generic or missing. E.g., 'shell' command param lacks timeout/output limits guidance; 'edit' oldString/newString lack examples of escaping or multi-line handling. Descriptions under 72 chars average (baseline 72).