Loading ToolBench
mimo-code has 33 tools with highly inconsistent quality. 12 tools lack visible descriptions entirely (GlobTool, GrepTool, WebFetchTool, WebSearchTool, TodoWriteTool, NotebookEditTool, PowerShellTool, ToolSearchTool, TaskStopTool, TaskOutputTool, SendMessageTool, ListMcpResourcesTool, ReadMcpResourceTool, SkillTool, TaskCreateTool, TaskUpdateTool, TaskListTool, LSPTool, MCPTool, ScheduleCronTool). Of the 13 with descriptions, most are under 100 chars and lack actionable context. Parameter descriptions are sparse: FileReadTool has good descriptions for offset/limit, but GlobTool, GrepTool, WebFetchTool lack any visible schema. BashTool and PowerShellTool expose command execution without clear safety guardrails or error recovery guidance. AgentTool spawns sub-agents but lacks documentation on resource limits, timeout behavior, or failure modes. No tools document output schemas. Error handling is minimal, most tools appear to return raw results without structured error classification or recovery hints. Naming is generally verb-first (good), but tools like 'EnterPlanMode' and 'ExitPlanMode' are vague about what 'plan mode' entails.
Spawn a sub-agent to handle a task autonomously. The sub-agent has access to the same tools and runs its own query loop until the task is complete.
Ask the user a clarifying question.
Execute a shell command and return stdout/stderr.
Send a brief structured message with optional file attachments.
Read or modify configuration settings programmatically.
进入计划模式。在此模式下只能读取文件和搜索，不能写入或执行命令。
Create and enter a git worktree for isolated work.
20 of 33 tools (61%) have no visible description in source code. This violates the critical pattern:tool-description requirement and prevents LLMs from determining when/why to select these tools.
No visible output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing what each tool returns. This violates pattern:tool and pattern:response-shaper.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
退出计划模式，提交计划供用户审批。
Exit and optionally remove a git worktree.
Replace an exact string in a file. By default replaces only the first occurrence.
Read the contents of a file. Supports reading with line offset and limit.
Write content to a file. Creates parent directories if needed.
BashTool and PowerShellTool expose arbitrary command execution without documented safety guardrails, timeout enforcement, or error recovery guidance. No mention of sandboxing, input validation, or injection prevention. Violates pattern:tool-gateway and pattern:recovery-guide.
AgentTool spawns sub-agents with no documented resource limits, timeout behavior, failure modes, or permission inheritance. Agents can recursively spawn agents without bounds, risking runaway execution and context exhaustion.
Parameter descriptions are sparse or missing. GlobTool, GrepTool, WebFetchTool, and 17 others have no visible schema definitions. Even tools with schemas (FileReadTool, FileEditTool) lack descriptions for all parameters. Violates pattern:tool-description and pattern:constrained-input.