Loading ToolBench
Skillmgmnt MCP server — lets your own agent query the hub as a native tool. Runs on your machine (stdio), talks to the hub over HTTP. Read-only access to skill inventory, search, health findings, safety vetting, and usage events.
7 tools with complete input schemas and descriptions, but descriptions are overly technical and Chinese-language, making them less useful for LLM selection. Tool names lack action verbs (overview, search_skills, skill_detail are acceptable; health_findings, safety_findings, remediation_plan are vague). Parameters are typed but descriptions are sparse. No output schemas documented. No error handling guidance. No parameter constraints (enums, ranges). Composition is reasonable (single responsibility per tool), but descriptions don't explain WHEN to use each tool or what downstream tools need.
治理发现：散落实体、仓库自带、断链、无法判定的软链、未索引、描述过短、正文超标、重复副本、闲置技能等。不带参数时只返回各节标题与条数；传 section 取该节明细。
Skillmgmnt 总览：库内技能数、已载入数、调用数、工具数、断链数、元数据常驻 token 数，以及上报的主机列表。回答「现在整体什么状况」。
结构化的治理方案：断链清单（含可安全执行的删除命令与重链命令）、散落实体清单（含替换为软链的建议命令，需人工确认后执行）、重复副本与安全高危摘要。命令在生成时已做 shell 引号处理；断链删除命令自带「执行时仍悬空才删」守卫。执行前向用户复述要做什么。
确定性安全审查结果（凭据与隐私 / 外发与下载 / 执行与提权）。命中是信号不是判决——每条带规则、文件、行号与原因，供人判断。
按关键词搜索技能（匹配 id/名称/描述/分类/标签，多词为 AND）。返回精简记录；要完整信息用 skill_detail。q 留空则列出全部。
一个技能的完整记录：描述、来源、分类、正文行数、载入到了哪些工具/项目（含软链类型与目标）、调用统计、上游更新状态、安全命中。
某个技能的调用记录（时间、工具、项目）。注意：12 个工具中仅 3 个会产出可解析的调用日志，所以「没有记录」不等于「没被用过」。
Descriptions are technical Chinese text without English translation, making them inaccessible to English-speaking LLMs and reducing tool discoverability. Example: 'Skillmgmnt 总览' instead of 'Get an overview of the skill inventory: total skills, loaded count, invocation count, tools, broken links, and metadata token usage.'
No output schemas documented. LLMs cannot plan downstream calls or extract required fields. Example: search_skills returns '精简记录' (simplified records) but structure is undefined.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Parameter descriptions are missing or minimal. 'q' in search_skills has description '关键词，空格分隔' (keywords, space-separated) but no guidance on matching behavior, case sensitivity, or max length. 'section' in health_findings lists examples but no explanation of what each section contains.
Tool names lack action verbs or are vague. 'health_findings', 'safety_findings', 'remediation_plan' do not clearly indicate what action is taken. Better: 'list_health_issues', 'scan_security_risks', 'generate_remediation_plan'.
No error handling guidance. If a skill key is not found in skill_detail, or if remediation_plan has no issues to fix, responses should guide the LLM on next steps (e.g., 'Try search_skills() to find the correct key').