@myagentroam/server 0.13.4 → 0.13.6

package/web/docs/assets/en-US_administration_license.md.jmAMVJjC.js

@@ -1 +0,0 @@
-import{_ as t,o as n,c as i,a3 as a}from"./chunks/framework.BHjrhdiC.js";const p=JSON.parse('{"title":"Community and Enterprise Licenses","description":"","frontmatter":{"title":"Community and Enterprise Licenses","lastUpdated":"2026-11-10T00:00:00.000Z"},"headers":[],"relativePath":"en-US/administration/license.md","filePath":"en-US/administration/license.md"}'),r={name:"en-US/administration/license.md"};function o(s,e,d,c,l,u){return n(),i("div",null,[...e[0]||(e[0]=[a('<h1 id="community-and-enterprise-licenses" tabindex="-1">Community and Enterprise Licenses <a class="header-anchor" href="#community-and-enterprise-licenses" aria-label="Permalink to &quot;Community and Enterprise Licenses&quot;">​</a></h1><p>Community Edition provides the default single-user experience without a limit on registered Nodes. An Enterprise license enables organizations, additional users, resource sharing, SSO, and audit management.</p><p>A System Administrator opens <strong>Admin → System Information</strong>, copies the request code, and sends it privately to the issuer. Paste the complete returned License document into <strong>License Authorization</strong>, import it, and verify enterprise name and expiry before creating enterprise resources.</p><p>An expired or frozen license restricts enterprise management writes while preserving configured resources and normal work according to the displayed state. Import a valid replacement to restore management. Treat request codes and License documents as administrative data and do not publish them.</p><p>Formal deployments use the same program and image set; activation does not replace the installation. Both formal license states require the real host UUID; containers must mount it read-only, and startup fails if it is unavailable or invalid. The local-only <a href="./../getting-started/npm-quick-start">npm Quick Start</a> uses fixed Community capabilities without a UUID and cannot be activated. License capabilities do not replace user and resource authorization.</p><h2 id="why-does-formal-community-deployment-require-a-host-uuid" tabindex="-1">Why does formal Community deployment require a host UUID? <a class="header-anchor" href="#why-does-formal-community-deployment-require-a-host-uuid" aria-label="Permalink to &quot;Why does formal Community deployment require a host UUID?&quot;">​</a></h2><p>In formal deployments, Community and Enterprise use the same Server startup and host-identity mechanism. Server derives a host fingerprint locally from the real UUID; Enterprise License validation uses it to bind a License to that Server host. Community does not require an Enterprise activation request or an online license-validation service.</p><p>This host-identity check is not an analytics upload. License verification is local; to request Enterprise activation, an administrator deliberately sends the request code (which contains the host fingerprint) to the issuer. Do not publish that code. This does not imply that the entire product is offline: model providers, configured integrations, and other network features have their own data flows.</p><p>Containers must mount the real host UUID read-only. A VM must expose a valid UUID; missing or invalid identity blocks startup rather than falling back to a generated ID. Moving Server to a host with a different UUID requires a replacement Enterprise License bound to the new host. Community still requires the new host&#39;s valid UUID but no Enterprise activation. Community is single-user and does not limit registered Nodes.</p><h2 id="source-licensing-and-permitted-use" tabindex="-1">Source licensing and permitted use <a class="header-anchor" href="#source-licensing-and-permitted-use" aria-label="Permalink to &quot;Source licensing and permitted use&quot;">​</a></h2><p>The EE author and licensor is <strong>aCHAOS (<a href="mailto:achaosxyz@gmail.com" target="_blank" rel="noreferrer">achaosxyz@gmail.com</a>)</strong>. <code>ee/LICENSE</code> contains the full English MyAgentRoam Enterprise Source-Available License 1.0 (<code>MAR-ESAL-1.0</code>); the product root <code>LICENSING.md</code> explains its scope. It is not an OSI-approved open-source license. Independent Apache components and third-party materials retain their own licenses.</p><p>Community features are free for internal use, including a company&#39;s own commercial work and deployment on cloud infrastructure it controls. Enterprise features include organization and multi-user administration, enterprise roles and organizational permissions, shared-resource authorization, SSO, enterprise operation auditing and associated governance. Production use requires an Enterprise license, except for the limited frozen-state continuation specified in the terms; isolated non-production development and testing are also permitted.</p><p>An Enterprise license activates features, not redistribution, resale, third-party hosting, SaaS/API operation, OEM or white-label rights. Unactivated or activated status, free third-party access and source modifications do not create exceptions; there is no separate commercial or hosting authorization to purchase. Internal hosted conversations are not third-party hosting services. Your own independent code and output are not subject to the EE license merely because You use MAR. The full scope, exceptions and contribution terms are in <code>ee/LICENSE</code>.</p>',13)])])}const h=t(r,[["render",o]]);export{p as __pageData,h as default};

package/web/docs/assets/en-US_workbench_skills.md.C4xlXyjC.js

@@ -1 +0,0 @@
-import{_ as t,o as n,c as a,a3 as s}from"./chunks/framework.BHjrhdiC.js";const p=JSON.parse('{"title":"Favorite and Install Skills","description":"","frontmatter":{"title":"Favorite and Install Skills","lastUpdated":"2026-11-06T00:00:00.000Z"},"headers":[],"relativePath":"en-US/workbench/skills.md","filePath":"en-US/workbench/skills.md"}'),o={name:"en-US/workbench/skills.md"};function r(i,e,l,d,c,g){return n(),a("div",null,[...e[0]||(e[0]=[s('<h1 id="favorite-and-install-skills" tabindex="-1">Favorite and Install Skills <a class="header-anchor" href="#favorite-and-install-skills" aria-label="Permalink to &quot;Favorite and Install Skills&quot;">​</a></h1><p>Configurable and added catalog lists request 20 entries per page from the Server. Search, ordering and categories cover the complete visible catalog; category tags do not shrink to the current page or search results. Source definitions and the current target’s installed Skill/MCP entries are complete configuration snapshots and are shown without pagination. An installation remains visible if its catalog reference disappears, but only its existing permitted operations remain available; this does not grant reinstallation or changes to inherited configuration.</p><p>Open <strong>Admin &gt; Skill Configuration</strong> to browse Skill and MCP entries made available from enabled sources. Skill Configuration, Skill Sources, and Secrets are consecutive flat Admin entries; there is no separate top-level Skills or Skill Center entry.</p><p>Choose the Node and, when required, the Workspace before installing. Review version, source, and target status. Updating replaces the installed copy with the selected catalog version; reinstalling restores the same version. If files conflict, inspect the target before approving an overwrite.</p><p>Skill Configuration shows installed state and environment checks. Removing an installation does not remove the catalog favorite or its source. Authorized users administer external sources under <strong>Admin &gt; Skill Sources</strong> and configure referenced secrets under <strong>Admin &gt; Environment Secrets</strong>.</p><h2 id="skill-sources-and-installation" tabindex="-1">Skill sources and installation <a class="header-anchor" href="#skill-sources-and-installation" aria-label="Permalink to &quot;Skill sources and installation&quot;">​</a></h2><p>Skills provide reusable instructions; MCP servers connect Agents to tools and services. Node-level configuration applies to its Workspaces, while a same-named Workspace configuration takes precedence. In <strong>Skill Sources</strong>, choose an enabled source and add a Skill to the available personal, organization, or enterprise catalog. Check its source and version before adding it. Catalog updates do not automatically update installed copies: check for updates, confirm the new version, and install it on each intended target.</p><p>Repository sources require a recognizable Git revision. Public GitHub and compatible GitLab repositories can use HTTPS; SSH sources use <code>git@host:owner/repo.git</code> and the Server deployment account&#39;s SSH configuration. SSH Skill sources skip host fingerprint verification, so use only trusted networks and sources; see <a href="./../getting-started/server-installation">Server deployment</a>. Source addresses are checked against local and private targets. TLS verification is enabled by default; disabling it for a self-signed source is a separate, high-risk administrator setting and does not relax address checks.</p><p>Choose an online Node and optionally its Workspace in <strong>Skill Configuration</strong>. Check the environment, then install, update, or reinstall. Node installations use the service account&#39;s <code>~/.agents/skills</code>; Workspace installations maintain <code>.agents/skills</code> and <code>.claude/skills</code>, with managed local Git exclusions rather than changes to the tracked <code>.gitignore</code>. A Node or browser interruption does not automatically retry an installation. If confirmation times out, refresh the target&#39;s actual status before deciding whether to retry. Unknown installed versions can be reinstalled; inspect permission, target path, and environment errors before trying again.</p><p>To see a Workspace&#39;s shared skills, open the Workspace and choose <strong>Loaded Skills</strong> from the menu next to <strong>New Session</strong>. This list and the Composer&#39;s <code>/</code> skill list read only the Node account&#39;s <code>~/.agents/skills</code> and the Workspace&#39;s <code>.agents/skills</code>, not the <code>.claude/skills</code> compatibility directories. Entries show their source, and a same-named Workspace skill takes precedence. Installation and removal remain in <strong>Skill Configuration</strong>; removing a Workspace override can expose a same-named Node entry again.</p><h2 id="mcp-servers" tabindex="-1">MCP servers <a class="header-anchor" href="#mcp-servers" aria-label="Permalink to &quot;MCP servers&quot;">​</a></h2><p>In <strong>MCP Sources</strong>, search an enabled Registry, inspect the server and pinned version, and add it to the available catalog. In <strong>Skill Configuration</strong>, switch to the MCP tab, select the online Node or Workspace, and install the intended entry. Node installation uses the Node account&#39;s command environment: npm-based entries may download a pinned package via <code>npx</code>. MAR validates initialization and reads at least one tool before saving the configuration. Resolve network, authentication, package, or system dependency errors on that Node before retrying.</p><p>Changes affect new or reopened Agent sessions, not an already running session. Removing the configuration does not clear npm caches. Environment variables and request headers refer to names in <strong>Environment Secrets</strong>, not secret values; add a missing referenced secret before using the MCP. Use <strong>Loaded MCP</strong> beside <strong>New Session</strong> to inspect the effective Workspace and Node configuration.</p><p>For Playwright, add a pinned Registry version and install it on the intended target. Its default browser is headless, each Agent session has an isolated browser context, and browser state is not retained after the session. The Node must be able to run the package and access the target site. If tools are unavailable, verify the target and installation and start a new session.</p>',14)])])}const u=t(o,[["render",r]]);export{p as __pageData,u as default};

package/web/docs/assets/workbench_skills.md.rWZ_VuoB.js

@@ -1 +0,0 @@
-import{_ as l,o as a,c as i,a3 as o}from"./chunks/framework.BHjrhdiC.js";const P=JSON.parse('{"title":"添加与安装技能","description":"","frontmatter":{"title":"添加与安装技能","lastUpdated":"2026-11-06T00:00:00.000Z"},"headers":[],"relativePath":"workbench/skills.md","filePath":"workbench/skills.md"}'),r={name:"workbench/skills.md"};function t(s,e,c,d,n,p){return a(),i("div",null,[...e[0]||(e[0]=[o('<h1 id="查找与配置-skill、mcp" tabindex="-1">查找与配置 Skill、MCP <a class="header-anchor" href="#查找与配置-skill、mcp" aria-label="Permalink to &quot;查找与配置 Skill、MCP&quot;">​</a></h1><p>可配置目录和已加入目录每页向服务端读取 20 条，搜索、排序和分类覆盖完整可见目录；标签分类不随当前页或搜索结果缩小。来源定义和当前目标的已安装 Skill/MCP 是完整配置快照，直接全部展示，不分页。已安装项即使失去目录引用仍保留，但只能使用原有合法操作，不能借此重新安装或更改继承配置。</p><p>“管理”菜单中的“技能配置”同时提供 Skill 与 MCP 两类 Agent 扩展。Skill 为 Agent 提供可复用的操作说明和工作流；MCP Server 让 Agent 连接浏览器、业务系统或其他工具。两者都可以配置到节点或工作区：节点配置对该节点上的工作区生效，工作区同名配置优先。</p><p>技能配置直接展示当前用户可用的企业、组织和个人目录，并用于将目录项安装到节点用户目录或工作区。页面默认在“可配置技能”Tab 选择并安装 Skill，“当前目标中的技能”Tab 用于查看、更新、重新安装或删除目标目录中的 Skill。</p><p>打开 Workspace 后，可点击会话栏“新建会话”右侧的“更多”按钮，再选择“已加载技能”，只读查看该 Workspace 的公共技能。此列表和 Composer 的 <code>/</code> 技能列表只读取节点用户目录 <code>~/.agents/skills</code> 与 Workspace 的 <code>.agents/skills</code>，不读取 <code>.claude/skills</code> 兼容目录；列表标注来源，名称相同时以 Workspace 技能为准。安装、更新和删除仍在“管理 → 技能配置”中完成。</p><h2 id="管理来源与可用技能" tabindex="-1">管理来源与可用技能 <a class="header-anchor" href="#管理来源与可用技能" aria-label="Permalink to &quot;管理来源与可用技能&quot;">​</a></h2><ol><li>打开“管理 → 技能源管理”。下方默认进入“源内技能”Tab，在下拉框选择已启用的来源；页面会自动加载该来源的技能，skills.sh 默认显示缓存 24 小时的 Hot 热门榜，也可在下方输入名称或描述后搜索。“已加入的技能”Tab 用于维护已经加入可用目录的技能。</li><li>点击技能卡片的“加入可用目录”，选择个人、组织或企业范围；可同时启用星标推荐并添加最多 10 个 Tag 标签。标签下拉提供当前账号有权查看的已有标签，也可以直接输入新标签。加入后可按名称、描述和分类标签筛选，可查看详情，或编辑星标和标签。</li><li>已加入目录的技能显示当前 Commit 短版本和更新时间。“检查更新”只比较原仓库最新 Commit，发现新版本后再由用户确认更新；目录更新后，Node 或 Workspace 不会自动变化。</li></ol><p>星标技能会优先出现在技能配置的可配置目录中，Tag 用于说明技能用途。更新技能内容不会改变已有星标和 Tag。</p><p>仓库源必须具备可识别的 Git 版本，不接受直接下载地址。公开仓库可使用 HTTPS，支持 GitHub，以及使用标准接口的 GitLab.com/自托管 GitLab；SSH 仓库使用 <code>git@host:owner/repo.git</code> 格式。Compose 部署会把独立部署账号的完整 <code>.ssh</code> 目录只读提供给 Server，具体配置见<a href="./../reference/configuration">配置项参考</a>；连接时跳过主机指纹校验。仓库根目录或标准技能目录均可，系统会拒绝指向本机、私网或链路本地地址的来源。每个用户都可以添加自己的个人源，而且只有本人能查看和管理；企业或组织源需要相应管理权限。默认 <code>skills.sh</code> 源可停用但不能删除。</p><p>TLS 证书校验默认启用。企业管理员可在“管理 → 系统配置”打开高风险的关闭开关，以兼容自签名来源；该设置不会放宽来源地址校验。</p><h2 id="安装到节点或工作区" tabindex="-1">安装到节点或工作区 <a class="header-anchor" href="#安装到节点或工作区" aria-label="Permalink to &quot;安装到节点或工作区&quot;">​</a></h2><ol><li>打开“管理 → 技能配置”，直接在页面中选择在线节点以及安装位置；安装到工作区时再选择该节点下的工作区。</li><li>节点或工作区切换后会自动检查环境，也可手动点击“检查环境”，随后执行“安装”“更新”或“重新安装”。页面选择独立于工作台资源树，不会切换当前打开的工作区。</li><li>节点以 <code>~/.agents/skills</code> 为主目录，并在首次安装时检查或创建 <code>.claude/skills</code> 兼容链接；若已有目录不能安全链接，则维护兼容副本。</li><li>工作区同时维护 <code>.agents/skills</code> 和 <code>.claude/skills</code> 两份副本，并将受管目录写入本地 <code>.git/info/exclude</code>，不会修改版本化 <code>.gitignore</code>。</li></ol><p>安装由节点在后台下载、校验并写入目录；页面会等待完成后刷新状态，耗时较长时在对应卡片显示进度，无需重复点击或刷新。安装失败会释放当前目标，处理错误原因后可以再次安装；状态确认超时时应先刷新目录确认真实结果，再决定是否重试。浏览器断开或服务重启不会自动重复安装。</p><p>系统不会覆盖人工修改的判断结果；无法识别已安装版本时会显示“版本未知”，可以从技能配置重新安装。</p><p>技能配置提供名称/描述搜索和“全部 + 分类标签”筛选，技能卡片可查看详情并在页面内容区内独立滚动。手动“检查环境”会弹窗显示兼容模式、技能数量和待修复项；Workspace 环境检查发现 Git 本地排除规则异常时，可点击带工具图标的“修正环境”，确认后重新生成 MAR 维护的规则块，取消不会修改 Workspace。</p><h2 id="平台与失败处理" tabindex="-1">平台与失败处理 <a class="header-anchor" href="#平台与失败处理" aria-label="Permalink to &quot;平台与失败处理&quot;">​</a></h2><p>Linux 与 Windows 均支持节点和工作区技能；Windows 不能创建目录链接时会使用受管副本。节点离线、版本过旧、目录权限不足、技能根目录指向 Workspace 外部，或工作区 Git 排除文件不可写时，界面会禁用操作或显示明确错误。单个技能名称位置若为链接，重新安装会原子替换该链接且不会修改链接指向的外部内容；MAR 不会跟随或覆盖越界的技能根目录。</p><h2 id="使用-mcp-server" tabindex="-1">使用 MCP Server <a class="header-anchor" href="#使用-mcp-server" aria-label="Permalink to &quot;使用 MCP Server&quot;">​</a></h2><h3 id="将-mcp-server-加入中心" tabindex="-1">将 MCP Server 加入中心 <a class="header-anchor" href="#将-mcp-server-加入中心" aria-label="Permalink to &quot;将 MCP Server 加入中心&quot;">​</a></h3><ol><li>打开“管理 → MCP 源管理”。系统已提供官方 MCP Registry；也可以新增兼容 Registry，并选择个人、组织或企业范围。</li><li>选择一个已启用来源，按名称或说明搜索 MCP Server。可先点击“详情”核对 Server 名称、连接方式和固定版本，再点击“加入可用目录”，设置可见范围、推荐状态和标签；标签可以从当前已有标签中选择，也可以直接输入新标签。</li><li>在“已加入的 MCP”中可查看详情、编辑推荐和标签、检查 Registry 更新、确认更新或移除。更新可用目录不会自动改变已经配置到节点或工作区的版本。</li><li>返回“管理 → 技能配置”，切换到“MCP”Tab，即可浏览当前账号可用的 MCP Server；可按名称或说明搜索，也可使用“全部 + 标签”切换分类，并可在选择目标后直接安装。</li></ol><p>目录只保存固定版本，不使用会在未来自动变化的 <code>latest</code>。新增或升级版本不会自动改变节点与工作区中的现有配置，需要用户明确安装。</p><h3 id="配置到节点或工作区" tabindex="-1">配置到节点或工作区 <a class="header-anchor" href="#配置到节点或工作区" aria-label="Permalink to &quot;配置到节点或工作区&quot;">​</a></h3><ol><li>打开“管理 → 技能配置”，切换到“MCP”Tab。</li><li>选择“节点”或“工作区”，然后选择在线目标。节点配置适合多个工作区共用；仅某个项目需要的 MCP 应配置到对应工作区。目标离线、停用、版本不支持、工作区缺失或权限不足时，页面会保留具体原因，不再只显示笼统的不可用提示。</li><li>在“可配置 MCP”Tab 可查看详情并点击“安装”；已经安装旧版本时按钮显示“更新”。安装由目标 Node 使用其运行账号的本地命令环境执行：<code>npx</code> 可按用户自己的 npm 配置下载和缓存固定版本，MAR 每次都会完成 MCP 初始化和工具读取验证，并至少确认一个可用工具，耗时较长时卡片会显示进度。只有验证完成才会保存配置并提示成功；失败时可先在 Node 终端修正 npm、网络、鉴权或系统依赖后重试。安装成功后只影响随后新建或重新打开的 Agent 会话，不修改已经运行的会话；删除配置不会清理用户的 npm/npx 缓存。</li><li>在“当前目标中的 MCP”Tab 可查看详情、启用、停用、更新、重新安装或删除。工作区删除同名配置后，若节点仍配置了该 MCP，新会话会继续使用节点配置。</li><li>在节点工作台中，点击“新建会话”右侧的“…”并选择“已加载 MCP”，可以在创建会话前确认该 Workspace 最终启用的 MCP 及其来自 Workspace 还是 Node。被 Workspace 同名配置停用的 Node MCP 不会出现在这里。</li></ol><p>MCP Server 的环境变量和请求头只引用“环境密钥”中已有的环境变量名，页面和保存记录不会包含密钥值。引用的环境密钥不存在时，该 MCP 不会注入 Agent 会话，应先在“管理 → 环境密钥”补齐。</p><h3 id="使用-playwright" tabindex="-1">使用 Playwright <a class="header-anchor" href="#使用-playwright" aria-label="Permalink to &quot;使用 Playwright&quot;">​</a></h3><p>在官方 MCP Registry 中搜索 Playwright，将需要的固定版本加入可用目录并安装到目标后，Agent 即可在浏览器中打开页面、交互、截图和验证 Web 功能。Playwright 不是不可删除的内置目录项，可从可用目录、节点或工作区移除。默认配置具有以下边界：</p><ul><li>使用无头浏览器，不弹出本机浏览器窗口；</li><li>每个 Agent 会话使用独立浏览器上下文，不共享 Cookie、登录态或页面；</li><li>会话结束后不保留浏览状态；需要登录时，应在当前会话内完成；</li><li>Node 主机需要能够执行该 MCP Server 的固定 npm 包版本并访问待测页面。</li></ul><p>如果 Agent 无法调用 Playwright，先确认目标节点在线、Node 版本支持 MCP、Playwright 已安装到正确的节点或工作区，并新建一个会话重试。运行中的旧会话不会热加载刚安装的 MCP。</p>',28)])])}const k=l(r,[["render",t]]);export{P as __pageData,k as default};