项目中心化的 Agent Skills 管理工具(Web GUI / Electron 桌面 App / CLI)
This HTTP server exposes 43 REST endpoints as MCP tools, but lacks critical MCP-compliant tool definitions. Tools are inferred from HTTP routes rather than explicitly registered with JSON Schema input/output definitions. Descriptions exist but are minimal (avg ~40 chars, mostly in Chinese). No input schemas are visible in the source, parameters are documented in prose descriptions only. No output schemas documented. Error handling returns generic HTTP status codes without recovery guidance. Tool naming uses HTTP verbs (GET/POST/PATCH/DELETE) rather than action verbs (get_, create_, update_). No tool annotations (readOnlyHint, destructiveHint, idempotentHint). This is a REST API wrapper, not a properly structured MCP server.
删除 MCP 服务
删除指定项目
获取全部支持的 agent 列表,包含 id、displayName、detected 状态与 capabilities
获取 AI 模型配置(脱敏后的公开部分)
列出推荐库中的技能与 MCP 服务(支持分类与类型过滤)
列出推荐库分类
环境自检,返回版本号与诊断报告(与 ssw doctor 同一份报告)
获取全局(用户级)共享档案
列出 MCP 服务中央注册表
No explicit MCP tool registration with JSON Schema. Tools are inferred from HTTP routes; input/output schemas are not formally declared in MCP format. This violates the tool definition pattern and prevents LLMs from understanding parameter constraints and output structure.
Tool naming uses HTTP verbs (GET, POST, PATCH, DELETE) instead of action verbs (get_, create_, update_, delete_). LLMs cannot infer intent from HTTP method names. Rename tools to verb_noun format: 'list_projects', 'create_project', 'update_project', 'delete_project'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |
获取服务进程信息,包含 cwd(当前工作目录,用于新建项目预填路径)
获取长任务进度(git clone/pull 等)
列出全部项目
获取指定项目详情
列出技能库中全部技能
获取更新下载进度
获取更新检查状态
更新 AI 模型配置(provider、model、apiKey 等)
更新全局共享档案(skills、agents、applyMode)
更新项目配置(name、agents、skills、mcps、applyMode)
测试 AI 连接
从 GitHub 仓库获取 MCP 配置
在 GitHub 上搜索技能或 MCP 服务
应用全局共享档案
取消应用全局共享档案
添加或更新 MCP 服务
导出项目配置档案(含内嵌技能代码)
导入项目配置档案
创建新项目,需提供 name、path(缺省为服务进程 cwd)、agents(缺省空数组)、applyMode(缺省 symlink)
获取 AI 推荐技能(读本地库 + 联网搜 GitHub)
应用项目配置(物化技能集与 MCP 服务到各 agent)
绑定/更新项目 MCP 服务集
回滚项目配置到上一个快照
绑定/更新项目技能集
切换为当前项目并应用配置
取消应用项目配置(移除物化的技能与 MCP 条目)
导出技能代码为迁移码
从迁移码导入技能
初始化新技能(创建 SKILL.md 与目录结构)
从 GitHub 安装技能
从本地路径安装技能
卸载技能
检查更新
下载更新
Descriptions are minimal (avg ~40 chars) and mostly in Chinese. Many lack context for when/why to use the tool. Descriptions should be 50-200 chars, in English, and answer: What does it do? When to use it? What does it return?
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields without knowing response structure. Document return types for all tools, especially those returning lists or complex objects.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot distinguish safe reads from destructive writes. Annotate DELETE and PATCH endpoints with destructiveHint=true; GET endpoints with readOnlyHint=true.