Loading ToolBench
A Go-based MCP server implementing an agent system with tool execution, skill management, capability disclosure, and multi-modal AI integration (vision, search, memory archiving).
This server exhibits severe definition quality issues across nearly all dimensions. Tool names lack clear action verbs (e.g., 'SetExecutorStep', 'AbortExecutorStep' use PascalCase inconsistently). Descriptions are present but often vague or domain-specific without explaining WHEN/WHY to use tools. Input schemas exist but lack proper type definitions for many parameters, several tools have object properties without 'type' declarations. Critical security and design flaws: 'evolve_persona_config' is marked IRREVERSIBLE and modifies agent personality persistently, violating the principle that tools should be composable and reversible. 'archive_specific_rounds' and 'report_step_result' appear to be internal framework tools exposed as user-callable tools, violating separation of concerns. Output schemas are not documented. Error handling is absent, no guidance on recovery, retryability, or actionable error messages.
通过连接高级 AI(Qwen3-Max) 搜索引擎获取经过提炼的实时信息
Abort the current executor step execution.
Execute a skill or MCP tool as part of the agent execution plan. Input format: {"skill":"skill_name","args":[...]} for skills or {"_call":"mcp","name":"tool_name","params":{...}} for MCP tools.
按需主动触发：当你发现对话历史中某些特定的轮次（Round）不是特别重要可以当前内存中移除时使用。输入参数是轮次索引列表。第一轮对话为1。例如：{"rounds": [1, 3, 5]} 表示将第1、3、5轮对话移入长期记忆并从当前内存删除。
内核人格演化工具：用于持久化修改你的底层配置文件。你可以指定修改身份认同、认知偏好或情感反应。参数格式：{"target_field": "identity|cognitive_bias|emotional_reaction|prompt_fragment", "new_content": "具体的描述内容"}注意：这会永久改变你的性格基石，请谨慎操作。
获取 MCP Schema、外挂 SKILL 全文或内置技能参数说明。JSON：{"mcp_tools":[],"external_skills":[],"builtin_skills":[]}。 mcp_tools 可填 MCP server 名（如 filesystem，展开该服务全部工具）或公开工具名（如 filesystem__read_text_file）。可组合、可多项。
获取周围环境的视觉消息。输入应为包含 direction 的 JSON 字符串，direction 可选值为：'前'、'后'、'左'、'右'。例如：{"direction": "前"}
SetExecutorStep and AbortExecutorStep use PascalCase naming instead of snake_case verb_noun convention. Names do not clearly convey action (SetExecutorStep could mean 'configure', 'register', or 'execute'). Violates pattern:tool naming standard.
report_step_result and AbortExecutorStep have no visible input schemas in source code. These appear to be internal framework tools exposed as user-callable tools, violating separation of concerns.
evolve_persona_config is marked IRREVERSIBLE and modifies agent personality persistently. This violates idempotent-operation and confirmation-request patterns. Agents should not have tools that permanently alter their own behavior without explicit user consent and undo capability.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | 2026-07-28+ | v2 |
返回当前进程 AGENTS.md 第一层能力目录全文（与 system 文末注入内容一致）。用户问「有哪些能力/技能/MCP」时优先调用本工具，禁止自行编造未列出的能力。无参：Action Input: {}
Report the result of an executed step back to the executor. Used internally by the execution framework.
No tool has documented output schemas. LLMs cannot plan downstream calls or extract required fields without knowing what each tool returns. Violates pattern:tool and pattern:response-shaper.
Descriptions for report_step_result, AbortExecutorStep, and archive_specific_rounds are under 50 characters and lack context on WHEN/WHY to use them. Violates pattern:tool-description.