@dai_ming/plugin-deliverables 1.0.0

package/README.md

@@ -0,0 +1,214 @@
+# @dai_ming/plugin-deliverables
+
+OpenClaw 交付物插件 — 让 AI Agent 将生成的文件（文章、HTML 页面、多文件游戏/网站、图片等）自动上传到 OSS，并在会话消息中回显可点击的预览/下载链接。
+
+---
+
+## 插件包含什么
+
+| 文件 | 作用 |
+|------|------|
+| `mcp-servers/deliverables.js` | MCP Server 脚本：实现 `upload_deliverable` 工具，通过 HTTP 调用 claw-gateway API 上传文件 |
+| `skills/deliverables/SKILL.md` | OpenClaw Skill：强制文档产出走 `upload_deliverable`，规范写入路径和参数 |
+| `agents-rules/deliverables.md` | AGENTS.md 规则块：URL 回显规则 + 自动上传规则（硬约束，注入到每个 workspace 的 AGENTS.md） |
+| `openclaw-plugin.json` | 插件清单：声明 MCP server、skill、AGENTS 规则、openclaw.json 配置段 |
+
+---
+
+## 安装方式
+
+### 方式一：通过 claw-gateway Helm 部署（推荐）
+
+只需在 Helm values 里把插件加入 `installPlugins` 列表，gateway 和 Helm initContainer 会自动完成所有配置：
+
+```yaml
+# values.yaml（或 claw-gateway 管理界面的 Helm 参数）
+installPlugins:
+  - "@dai_ming/plugin-deliverables@1.0.0"
+```
+
+initContainer 执行顺序：
+1. **Phase 3**：`npm pack @dai_ming/plugin-deliverables@1.0.0` 下载 tarball → 解压到 `/data/extensions-extra/plugin-deliverables/`
+2. **Phase 3b**：在插件目录执行 `npm install --omit=dev`（本插件无运行时依赖，此步骤跳过）
+3. **Phase 3e**：读取 `openclaw-plugin.json` 清单，自动：
+   - 复制 `mcp-servers/deliverables.js` → `/data/mcp-servers/deliverables.js`
+   - 安装 `skills/deliverables/SKILL.md` → 所有 workspace 的 `skills/deliverables/SKILL.md`
+   - 注入 `agents-rules/deliverables.md` 内容到所有 workspace 的 `AGENTS.md`（幂等，按 marker 替换）
+
+> **env 变量**：MCP Server 需要以下变量（claw-gateway 在生成 `mcp.servers` 配置时会自动注入）：
+>
+> | 变量 | 说明 | 默认值 |
+> |------|------|--------|
+> | `CLAW_GATEWAY_URL` | gateway 内部地址 | `http://claw-gateway:8080` |
+> | `CLAW_GATEWAY_PUBLIC_URL` | gateway 公网地址（用于生成预览链接） | 同上 |
+> | `CLAW_GATEWAY_API_KEY` | API Key | `api-key-1` |
+
+### 方式二：手动安装（不使用 claw-gateway 自动部署）
+
+适用于自行管理 OpenClaw 容器的场景。
+
+**Step 1 — 下载并解压插件**
+
+```bash
+npm config set registry https://registry.npmmirror.com
+mkdir -p ~/.openclaw/extensions-extra/plugin-deliverables
+cd /tmp && npm pack @dai_ming/plugin-deliverables
+tar xzf openclaw-plugin-deliverables-*.tgz -C ~/.openclaw/extensions-extra/plugin-deliverables --strip-components=1
+```
+
+**Step 2 — 复制 MCP Server 脚本**
+
+```bash
+mkdir -p ~/.openclaw/mcp-servers
+cp ~/.openclaw/extensions-extra/plugin-deliverables/mcp-servers/deliverables.js \
+   ~/.openclaw/mcp-servers/deliverables.js
+```
+
+**Step 3 — 安装 Skill**
+
+```bash
+WORKSPACE=~/.openclaw/workspace   # 替换为实际 workspace 路径
+mkdir -p "$WORKSPACE/skills/deliverables"
+cp ~/.openclaw/extensions-extra/plugin-deliverables/skills/deliverables/SKILL.md \
+   "$WORKSPACE/skills/deliverables/SKILL.md"
+```
+
+**Step 4 — 注入 AGENTS.md 规则**
+
+```bash
+AGENTS="$WORKSPACE/AGENTS.md"
+RULES=~/.openclaw/extensions-extra/plugin-deliverables/agents-rules/deliverables.md
+
+if grep -q "DELIVERABLE_LINK_RULES_START" "$AGENTS" 2>/dev/null; then
+  # 已有旧规则，用 Node.js 按 marker 替换
+  node -e "
+    var fs=require('fs');
+    var a=fs.readFileSync('$AGENTS','utf8');
+    var r=fs.readFileSync('$RULES','utf8').trim();
+    var re=/<!--\s*DELIVERABLE_LINK_RULES_START\s*-->[\s\S]*?<!--\s*DELIVERABLE_AUTO_UPLOAD_RULES_END\s*-->/g;
+    fs.writeFileSync('$AGENTS', re.test(a) ? a.replace(re,r) : a+'\n\n'+r+'\n');
+  "
+else
+  # 首次注入
+  printf '\n\n' >> "$AGENTS"
+  cat "$RULES" >> "$AGENTS"
+fi
+```
+
+**Step 5 — 配置 openclaw.json**
+
+在 `~/.openclaw/openclaw.json` 的 `mcp.servers` 下添加：
+
+```json
+{
+  "mcp": {
+    "servers": {
+      "deliverables": {
+        "command": "node",
+        "args": ["/home/node/.openclaw/mcp-servers/deliverables.js"],
+        "env": {
+          "CLAW_GATEWAY_URL": "http://<your-gateway-host>:8080",
+          "CLAW_GATEWAY_PUBLIC_URL": "https://<public-domain>",
+          "CLAW_GATEWAY_API_KEY": "<your-api-key>"
+        }
+      }
+    }
+  },
+  "plugins": {
+    "entries": {
+      "deliverables": { "enabled": true }
+    }
+  }
+}
+```
+
+---
+
+## AGENTS.md 规则说明
+
+插件向每个 workspace 的 `AGENTS.md` 注入两段硬约束规则：
+
+| 规则段 | Marker | 作用 |
+|--------|--------|------|
+| URL 回显规则 | `DELIVERABLE_LINK_RULES_START` … `DELIVERABLE_LINK_RULES_END` | 强制 Agent 上传后在消息中输出可点击的 Markdown 链接 |
+| 自动上传规则 | `DELIVERABLE_AUTO_UPLOAD_RULES_START` … `DELIVERABLE_AUTO_UPLOAD_RULES_END` | 当用户要求生成文件时默认自动上传，写入 `output/` 目录 |
+
+规则使用 HTML 注释 Marker 包裹，initContainer 和 Guardian 进程每次同步时都会幂等替换，不会产生重复块。
+
+---
+
+## Skill 说明
+
+`skills/deliverables/SKILL.md` 是 OpenClaw Skill，被注入到 workspace 后由 Agent 在工具调用前读取。它规定：
+
+- 始终使用会话中实际暴露的工具名（`deliverables__upload_deliverable`）
+- 生成文件统一写入 `output/` 子目录
+- 多文件（游戏/网站）优先用 `type=game` + `files[]`，不要提前打 zip
+- 上传成功后回复必须附带 Markdown 格式的预览/下载链接
+
+---
+
+## MCP Server 说明
+
+`mcp-servers/deliverables.js` 是一个符合 [MCP 协议（2024-11-05）](https://modelcontextprotocol.io/) 的 Node.js stdio server，暴露单个工具：
+
+### `upload_deliverable`
+
+| 参数 | 类型 | 必填 | 说明 |
+|------|------|------|------|
+| `resource_id` | string | ✅ | 当前会话唯一 ID |
+| `type` | enum | ✅ | `article` / `game` / `zip` / `image` / `video` / `ppt` / `link` |
+| `file_name` | string | ✅ | 用户可见文件名（含扩展名）或目录名 |
+| `group_id` | string | — | 群聊 ID |
+| `user_id` | string | — | 用户 ID |
+| `content_text` | string | — | 文本内容（单文件场景） |
+| `content_base64` | string | — | Base64 编码的二进制内容（单文件场景） |
+| `files` | array | — | 多文件列表（游戏/站点场景，优先使用） |
+
+返回值包含 `preview_url`、`download_url`、`reply_markdown`（可直接附在回复消息中）。
+
+---
+
+## 版本与发布
+
+### 在 claw-gateway 仓库内发布
+
+插件源码位于 `package/` 目录，与 claw-gateway 主仓库同步维护。更新流程：
+
+```bash
+# 1. 修改 package/package.json 中的 version
+# 2. 同步更新 package/mcp-servers/deliverables.js（如有变更）
+# 3. 打包并发布到 npmmirror 兼容的私有或公开 registry
+cd package
+npm publish --registry https://registry.npmmirror.com
+```
+
+### 版本对齐建议
+
+| claw-gateway 版本 | plugin 版本 |
+|-------------------|-------------|
+| 当前              | 1.0.0       |
+
+---
+
+## 常见问题
+
+**Q: 插件安装后 Agent 没有 `upload_deliverable` 工具怎么办？**
+
+检查以下几点：
+1. `openclaw.json` 的 `mcp.servers.deliverables` 是否存在
+2. MCP Server 是否在运行：`kubectl exec <pod> -- ls /home/node/.openclaw/mcp-servers/`
+3. Agent 的 `tools.allow` 是否包含 `deliverables__upload_deliverable`
+
+**Q: 上传后没有返回链接怎么办？**
+
+确认 `CLAW_GATEWAY_URL` 和 `CLAW_GATEWAY_API_KEY` 注入正确，可在容器内测试：
+
+```bash
+curl -H "X-API-Key: $CLAW_GATEWAY_API_KEY" $CLAW_GATEWAY_URL/healthz
+```
+
+**Q: claw-gateway 已经内置了 deliverables，会和插件冲突吗？**
+
+不会冲突。claw-gateway 通过 ConfigMap 注入的 MCP 脚本、skill、AGENTS 规则与插件内容完全一致。
+Phase 3e（插件注入）在前，ConfigMap 阶段（Phase 4/4c/4d）在后；ConfigMap 的内容会覆盖插件安装的内容，两者互为备份。

package/agents-rules/deliverables.md

@@ -0,0 +1,77 @@
+<!-- DELIVERABLE_LINK_RULES_START -->
+## Deliverables -- URL Echo Rule (HARD CONSTRAINT)
+
+When you call the deliverables upload tool, your final assistant message MUST include clickable URLs in Markdown link format (short label + full URL target).
+
+Tool name note:
+
+- Always call the exact tool name exposed in the current session.
+- In current runtime, the deliverables tool is usually:
+  - `deliverables__upload_deliverable`
+- If notes mention bare names like `upload_deliverable`, do **not** guess. Use the actual namespaced tool that is exposed to you.
+
+### Required output behavior
+
+0. Your final assistant message MUST start with 1-2 short sentences in your own words, briefly describing what you produced for the user.
+1. The intro must be based on the actual content/request, not a fixed boilerplate like `交付物已上传成功，可直接在线预览或下载。`
+2. If tool result contains `reply_markdown`, append that field verbatim after your intro on the following lines. Do not rewrite the links inside it.
+3. If tool result has `preview_url`, include one line:
+   `预览链接：[点击预览](<full_url>)`
+4. If tool result has `download_url`, include one line:
+   `下载链接：[点击下载](<full_url>)`
+5. For multi-file/game deliverables, if tool result already formats the second line as
+   `文件列表：[查看目录](<full_url>)`
+   then keep that exact label and URL. Do not rewrite it back to a raw long link.
+6. Keep URL target value exactly from tool result (no shortening, no masking, no redirect rewrite).
+7. Use short link labels (`点击预览` / `点击下载` / `查看目录`) to avoid exposing long raw URLs.
+8. Do not say "已上传/已完成" without links.
+9. Do not output naked long URLs outside Markdown link syntax.
+10. Keep the intro concise; do not paste the full file content, saved-path text, or a long summary after the Markdown links.
+
+### Forbidden output behavior
+
+- "可点击预览链接查看" but no actual URL
+- Replacing URL target with a non-original URL
+- Omitting both links when tool already returned links
+- Outputting only naked long URL without Markdown link label
+- Replacing `文件列表：[查看目录](...)` with a naked URL or a zip description
+- Using a generic boilerplate intro that does not mention the actual deliverable content
+<!-- DELIVERABLE_LINK_RULES_END -->
+
+<!-- DELIVERABLE_AUTO_UPLOAD_RULES_START -->
+## Deliverables -- Auto Upload for Document Requests (HARD CONSTRAINT)
+
+When user asks to write/generate/create a file, document, article, report, HTML, PPT, plan, guide, markdown, introduction, biography, strategy, summary, or similar artifact, you MUST upload it as a deliverable by default.
+
+Tool name note:
+
+- Always call the exact tool name exposed in the current session.
+- In current runtime, the upload tool is `deliverables__upload_deliverable`.
+- Do not call a guessed bare tool name if the actual tool list shows a namespaced one.
+
+### Required behavior
+
+1. Create content under current agent workspace `output/` directory first (for example `output/report.md`), then call the deliverables upload tool exposed in this session.
+2. If you need to use the `write` tool, the target path MUST be inside `output/`. Never write deliverable files to the workspace root.
+3. If you already wrote the file to the wrong place, move/copy it into `output/` before finalizing the task and before describing the saved path to the user.
+4. If format is HTML, prefer `type=article` and file name ends with `.html`.
+5. If format is markdown/text, use `type=article` and `.md`/`.txt`.
+6. For multi-file deliverables (game/site), you MUST prefer `type=game` with `files[]`, keep files as a folder structure, and do NOT zip before upload unless the user explicitly asks for a zip package.
+7. Static multi-file game/site deliverables SHOULD include a root `index.html` so the preview link can open the homepage directly.
+8. If a generated project requires starting a separate backend service, custom port, database, or long-running process to work, do NOT pretend the deliverables preview can run it. Tell the user that deliverables preview only supports static output, and that runtime projects need deployment/ingress instead.
+9. After successful upload, the final assistant message MUST start with 1-2 short sentences in your own words, briefly describing what you generated for the user.
+10. The intro must be based on the actual deliverable content/request, not a fixed sentence like `交付物已上传成功，可直接在线预览或下载。`
+11. If tool result contains `reply_markdown`, append that field verbatim after your intro on the following lines.
+12. Otherwise final reply MUST include Markdown links (short label + full URL target):
+   `预览链接：[点击预览](<full_url>)`
+   `下载链接：[点击下载](<full_url>)`
+13. For multi-file/game deliverables, if the tool gives directory-style output, the second line may be:
+   `文件列表：[查看目录](<full_url>)`
+   Keep that format instead of forcing a zip link.
+14. Do NOT only say "已保存到工作空间".
+15. Do NOT append workspace path or a raw URL block after the Markdown links.
+
+### Exception
+
+- Only skip upload when user explicitly says: "不要上传交付物" / "只保存在工作区".
+<!-- DELIVERABLE_AUTO_UPLOAD_RULES_END -->

package/mcp-servers/deliverables.js

@@ -0,0 +1,295 @@
+#!/usr/bin/env node
+"use strict";
+
+/**
+ * deliverables MCP Server
+ *
+ * Provides tools for uploading AI-generated content (articles, games, images, etc.)
+ * to OSS via the claw-gateway API and returning shareable URLs.
+ *
+ * Env vars (injected by helm_deployer.go):
+ *   CLAW_GATEWAY_URL     — base URL of claw-gateway, e.g. http://claw-gateway:8080
+ *   CLAW_GATEWAY_API_KEY / OPENCLAW_GATEWAY_API_KEY — API key with access to /openclaw-gateway/be/*
+ */
+
+var http  = require("http");
+var https = require("https");
+
+var GATEWAY_URL    = (process.env.CLAW_GATEWAY_URL        || "http://claw-gateway:8080").replace(/\/$/, "");
+var GATEWAY_PUBLIC = (process.env.CLAW_GATEWAY_PUBLIC_URL || GATEWAY_URL).replace(/\/$/, "");
+// Some existing pods were created without CLAW_GATEWAY_API_KEY injected.
+// Keep env-driven behavior first, but provide a dev-compatible fallback to avoid
+// breaking deliverable uploads during rolling migration.
+var API_KEY        = process.env.CLAW_GATEWAY_API_KEY || process.env.OPENCLAW_GATEWAY_API_KEY || "api-key-1";
+
+var TOOL_DEFS = [
+  {
+    name: "upload_deliverable",
+    description: [
+      "将 AI 生成的内容（文章、游戏、图片等）上传为交付物，返回可分享的下载/预览链接。",
+      "单文件交付物：提供 content_text 或 content_base64。",
+      "多文件交付物（网页游戏/静态站点等）：必须优先提供 files 列表，每项包含 name（相对路径）和 content_text 或 content_base64，不要先打 zip。",
+      "静态多文件预览建议在根目录提供 index.html；需要单独启动端口/后端服务的项目不属于交付物预览范围，应走部署流程。",
+      "返回的 download_url / preview_url 为长期可用链接。"
+    ].join(" "),
+    inputSchema: {
+      type: "object",
+      properties: {
+        resource_id: {
+          type: "string",
+          description: "当前聊天框/会话的唯一 ID（必填）。对应消息上下文中的 resource_id 字段，直接使用该值即可。"
+        },
+        group_id: {
+          type: "string",
+          description: "当前会话所属的群聊 ID（可选）"
+        },
+        user_id: {
+          type: "string",
+          description: "请求交付物的用户 ID（可选）"
+        },
+        type: {
+          type: "string",
+          enum: ["article", "game", "zip", "image", "video", "ppt", "link"],
+          description: "交付物类型：article=文章/文档，game=静态多文件网页/游戏（优先使用该类型并传 files，除非用户明确要 zip），zip=通用压缩包，image=图片，video=视频，ppt=演示文稿，link=外部链接（无需上传内容，fileName 填 URL）"
+        },
+        file_name: {
+          type: "string",
+          description: "用户可见的文件名，例如 adventure-game 或 report.md。多文件交付时这里应是目录名/项目名，不要写成 .zip，除非用户明确要求压缩包。"
+        },
+        content_text: {
+          type: "string",
+          description: "文本内容（文章、HTML、Markdown 等），单文件时使用"
+        },
+        content_base64: {
+          type: "string",
+          description: "Base64 编码的二进制内容（图片、zip 等），单文件时使用"
+        },
+        files: {
+          type: "array",
+          description: "多文件列表（游戏/静态站点场景）必须优先使用，服务端会保留目录结构，不再打 zip。静态预览场景建议包含根目录 index.html。",
+          items: {
+            type: "object",
+            properties: {
+              name:           { type: "string", description: "文件在交付物目录内的相对路径，例如 index.html 或 assets/main.js" },
+              content_text:   { type: "string", description: "文本内容" },
+              content_base64: { type: "string", description: "Base64 二进制内容" }
+            },
+            required: ["name"]
+          }
+        }
+      },
+      required: ["resource_id", "type", "file_name"]
+    }
+  }
+];
+
+// ─── HTTP helpers ─────────────────────────────────────────────────────────────
+
+function parseURL(rawURL) {
+  try {
+    return new URL(rawURL);
+  } catch(e) {
+    return null;
+  }
+}
+
+function httpRequest(method, path, body) {
+  return new Promise(function(resolve, reject) {
+    var parsed = parseURL(GATEWAY_URL + path);
+    if (!parsed) {
+      return reject(new Error("invalid gateway URL: " + GATEWAY_URL + path));
+    }
+    var isHTTPS = parsed.protocol === "https:";
+    var transport = isHTTPS ? https : http;
+    var bodyStr = body ? JSON.stringify(body) : "";
+    var options = {
+      hostname: parsed.hostname,
+      port:     parsed.port || (isHTTPS ? 443 : 80),
+      path:     parsed.pathname + (parsed.search || ""),
+      method:   method,
+      headers: {
+        "Content-Type":  "application/json",
+        "X-API-Key":     API_KEY,
+        "Content-Length": Buffer.byteLength(bodyStr)
+      }
+    };
+    var req = transport.request(options, function(res) {
+      var chunks = [];
+      var traceID = res.headers["x-trace-id"] || "";
+      res.on("data", function(c) { chunks.push(c); });
+      res.on("end", function() {
+        var text = Buffer.concat(chunks).toString();
+        try {
+          var obj = JSON.parse(text);
+          if (res.statusCode >= 400) {
+            reject(new Error(formatGatewayError(res.statusCode, obj.message || text, traceID)));
+          } else {
+            resolve({ body: obj, traceID: traceID });
+          }
+        } catch(e) {
+          reject(new Error(formatGatewayError(res.statusCode, "non-JSON response: " + text.slice(0, 200), traceID)));
+        }
+      });
+    });
+    req.on("error", reject);
+    if (bodyStr) req.write(bodyStr);
+    req.end();
+  });
+}
+
+function formatGatewayError(statusCode, message, traceID) {
+  var msg = "gateway " + statusCode + ": " + message;
+  if (traceID) {
+    msg += " (trace_id=" + traceID + ")";
+  }
+  return msg;
+}
+
+function buildReplyMarkdown(opts) {
+  var previewURL = opts.previewURL || "";
+  var downloadURL = opts.downloadURL || "";
+  var deliverableType = opts.type || "";
+  var isDirectory = !!opts.isDirectory;
+  var lines = [];
+  if (previewURL) {
+    lines.push("预览链接：[点击预览](" + previewURL + ")");
+  }
+  if (downloadURL) {
+    if (isDirectory || deliverableType === "game") {
+      lines.push("文件列表：[查看目录](" + downloadURL + ")");
+    } else {
+      lines.push("下载链接：[点击下载](" + downloadURL + ")");
+    }
+  }
+  if (lines.length === 0 && !previewURL && !downloadURL) {
+    return "交付物已上传成功。";
+  }
+  return lines.join("\n");
+}
+
+// ─── Tool implementations ─────────────────────────────────────────────────────
+
+function uploadDeliverable(args) {
+  var body = {
+    resourceId: args.resource_id,
+    groupId:    args.group_id,
+    userId:     args.user_id,
+    type:       args.type,
+    fileName:   args.file_name,
+    release:    ""  // overwritten below after release name derivation
+  };
+
+  if (args.files && args.files.length > 0) {
+    body.files = args.files.map(function(f) {
+      return {
+        name:          f.name,
+        contentText:   f.content_text   || "",
+        contentBase64: f.content_base64 || ""
+      };
+    });
+  } else {
+    body.contentText   = args.content_text   || "";
+    body.contentBase64 = args.content_base64 || "";
+  }
+
+  // Derive release name:
+  // 1) botID (runtime env, e.g. "user-xxx") is preferred
+  // 2) OPENCLAW_RELEASE / BOT_ID (compat)
+  // 3) fallback to stripping "oc-" prefix from OPENCLAW_GATEWAY_TOKEN
+  var release = process.env.botID || process.env.OPENCLAW_RELEASE || process.env.BOT_ID || "";
+  if (!release) {
+    var tok = process.env.OPENCLAW_GATEWAY_TOKEN || "";
+    if (tok.indexOf("oc-") === 0) release = tok.slice(3);
+  }
+  body.release = release;
+
+  return httpRequest("POST", "/openclaw-gateway/be/deliverables", body).then(function(resp) {
+    var d = resp.body.data || resp.body;
+    // Prefer backend-aware previewUrl returned by gateway (OSS/output/link).
+    // Fallback to legacy gateway preview endpoint for compatibility.
+    var previewURL = d.previewUrl || (GATEWAY_PUBLIC + "/openclaw-gateway/preview/" + d.uuid);
+    var isDirectory = !!(args.files && args.files.length > 0);
+    if (!isDirectory && d.downloadUrl && previewURL && d.downloadUrl !== previewURL && /(?:\?|&)list=1(?:&|$)/.test(d.downloadUrl)) {
+      isDirectory = true;
+    }
+    var replyMarkdown = buildReplyMarkdown({
+      previewURL: previewURL,
+      downloadURL: d.downloadUrl,
+      type: args.type,
+      isDirectory: isDirectory
+    });
+    return {
+      uuid:         d.uuid,
+      backend:      d.backend || "",
+      download_url: d.downloadUrl,
+      preview_url:  previewURL,
+      expire_at:    d.expireAt,
+      reply_markdown: replyMarkdown,
+      trace_id:     resp.traceID || "",
+      message: replyMarkdown
+    };
+  });
+}
+
+// ─── MCP JSON-RPC loop ────────────────────────────────────────────────────────
+
+function callTool(name, args) {
+  switch (name) {
+    case "upload_deliverable":  return uploadDeliverable(args);
+    default: return Promise.reject(new Error("unknown tool: " + name));
+  }
+}
+
+function send(obj) {
+  process.stdout.write(JSON.stringify(obj) + "\n");
+}
+
+function handleMessage(msg) {
+  var id = msg.id, method = msg.method, params = msg.params || {};
+
+  switch (method) {
+    case "initialize":
+      send({ jsonrpc: "2.0", id: id, result: {
+        protocolVersion: "2024-11-05",
+        capabilities: { tools: {} },
+        serverInfo: { name: "deliverables", version: "1.0.0" }
+      }});
+      return Promise.resolve();
+
+    case "notifications/initialized":
+      return Promise.resolve();
+
+    case "tools/list":
+      send({ jsonrpc: "2.0", id: id, result: { tools: TOOL_DEFS } });
+      return Promise.resolve();
+
+    case "tools/call":
+      return callTool(params.name, params.arguments || {}).then(
+        function(result) {
+          send({ jsonrpc: "2.0", id: id, result: {
+            content: [{ type: "text", text: JSON.stringify(result, null, 2) }]
+          }});
+        },
+        function(err) {
+          send({ jsonrpc: "2.0", id: id, result: {
+            content: [{ type: "text", text: "Error: " + err.message }],
+            isError: true
+          }});
+        }
+      );
+
+    default:
+      send({ jsonrpc: "2.0", id: id, error: { code: -32601, message: "Method not found: " + method } });
+      return Promise.resolve();
+  }
+}
+
+var rl = require("readline").createInterface({ input: process.stdin, terminal: false });
+rl.on("line", function(line) {
+  if (!line.trim()) return;
+  var msg;
+  try { msg = JSON.parse(line); } catch(e) { return; }
+  handleMessage(msg).catch(function(err) {
+    process.stderr.write("[deliverables] unhandled error: " + err.message + "\n");
+  });
+});

package/openclaw-plugin.json

@@ -0,0 +1,33 @@
+{
+  "name": "deliverables",
+  "version": "1.0.0",
+  "npm_package": "@dai_ming/plugin-deliverables",
+  "description": "Deliverables plugin: MCP server + skill + AGENTS rules for AI-generated file uploads",
+  "mcp_servers": {
+    "deliverables": {
+      "script": "mcp-servers/deliverables.js",
+      "command": "node",
+      "env": {
+        "CLAW_GATEWAY_URL": "${CLAW_GATEWAY_URL}",
+        "CLAW_GATEWAY_PUBLIC_URL": "${CLAW_GATEWAY_PUBLIC_URL}",
+        "CLAW_GATEWAY_API_KEY": "${CLAW_GATEWAY_API_KEY}",
+        "OPENCLAW_GATEWAY_API_KEY": "${OPENCLAW_GATEWAY_API_KEY}"
+      }
+    }
+  },
+  "skills": {
+    "deliverables": "skills/deliverables/SKILL.md"
+  },
+  "agents_rules": {
+    "file": "agents-rules/deliverables.md",
+    "start_marker": "DELIVERABLE_LINK_RULES_START",
+    "end_marker": "DELIVERABLE_AUTO_UPLOAD_RULES_END"
+  },
+  "openclaw_config": {
+    "plugins": {
+      "entries": {
+        "deliverables": { "enabled": true }
+      }
+    }
+  }
+}

package/package.json

@@ -0,0 +1,21 @@
+{
+  "name": "@dai_ming/plugin-deliverables",
+  "version": "1.0.0",
+  "description": "OpenClaw deliverables plugin — upload AI-generated files to OSS and return shareable preview/download links",
+  "keywords": [
+    "openclaw",
+    "plugin",
+    "deliverables",
+    "mcp"
+  ],
+  "license": "MIT",
+  "files": [
+    "openclaw-plugin.json",
+    "mcp-servers/",
+    "skills/",
+    "agents-rules/"
+  ],
+  "engines": {
+    "node": ">=16"
+  }
+}

package/skills/deliverables/SKILL.md

@@ -0,0 +1,69 @@
+---
+name: deliverables
+description: 上传AI生成的文件到交付物系统，返回可分享的预览链接
+---
+
+# 文件交付规则（强制）
+
+当你生成了文件内容（文章、文档、介绍、攻略、HTML页面、Markdown、游戏、图片等），**必须**调用交付物上传工具，不得把文件内容直接粘贴在消息里。
+
+## 工具名说明（强制）
+
+- 始终调用当前会话里真实暴露出来的工具名。
+- 当前运行时里交付物工具就是：
+  - `deliverables__upload_deliverable`
+- 如果你在说明文字里同时看到了裸名 `upload_deliverable`，不要猜测；优先使用工具列表里实际出现的 namespaced 名称。
+
+## 工作区文件路径（强制）
+
+- 生成的文件默认写入当前 agent 工作区的 `output/` 子目录。
+- 不要直接写到工作区根目录（如当前 workspace 根目录下的 `*.md`），应写到当前 workspace 的 `output/*.md`。
+- 多文件内容（游戏/网站）写到 `output/<目录名>/...`，保持目录结构。
+- 如果你使用 `write` 工具创建交付物文件，目标路径必须在 `output/` 下。
+- 如果你已经误写到工作区根目录，必须先移动或复制到 `output/`，再继续上传交付物并向用户汇报。
+
+## 调用前必须准备的参数（全部必填）
+
+从当前会话上下文提取以下值，**调用时一次性填完，不要留空**：
+
+| 参数 | 取值来源 | 示例 |
+|------|---------|------|
+| `resource_id` | 消息元数据中的 `resource_id` 字段 | `user_xxx_lobster_yyy` |
+| `group_id` | 消息元数据中的 `group_id` 或 `conversation_id` | `group_abc123` |
+| `user_id` | 消息元数据中的 `sender_id` 或 `owner_id` | `cbb0fab9...` |
+| `type` | 根据内容选择：`article`/`game`/`image`/`video`/`ppt`/`zip`/`link` | `article` |
+| `file_name` | 有意义的文件名，含扩展名 | `report-2026.html` |
+| `content_text` | 文件的完整文本内容（HTML/Markdown等） | `<html>...</html>` |
+
+> **直接对话（direct chat）时**：`group_id` 填 `conversation_id`，`user_id` 填 `owner_id`。
+
+## 多文件（游戏）
+
+type 为 `game` 时，使用 `files` 数组替代 `content_text`：
+```
+files: [
+  { name: "index.html", content_text: "..." },
+  { name: "style.css",  content_text: "..." }
+]
+```
+
+强制要求：
+
+- 多文件网页/游戏默认按“目录”发布，不要先打 zip。
+- 除非用户明确要求压缩包，否则必须优先用 `type=game` + `files`。
+- 静态网页/游戏应尽量在根目录提供 `index.html`，这样预览链接可以直接打开主页。
+- 如果内容需要启动独立后端服务、端口、数据库或长运行进程，交付物预览本身无法运行它；这类场景应改走部署/ingress，不要伪装成“可直接预览”的交付物。
+
+## 上传成功后
+
+优先规则：
+- 最终消息必须先用你自己的话写 1-2 句简短介绍，粗略说明你生成了什么内容、包含哪些重点。
+- 这句介绍必须基于实际产物内容，不要使用固定模板，例如：`交付物已上传成功，可直接在线预览或下载。`
+- 如果工具结果里有 `reply_markdown`，把它原样放在这句介绍后面，不要改写其中的链接。
+
+否则只发给用户：
+- 预览链接（`preview_url`，必须用 Markdown 链接格式）
+- 下载链接（`download_url`，必须用 Markdown 链接格式）
+- 多文件/游戏目录场景下，第二行也可以是 `文件列表：[查看目录](...)`，不要改成裸 URL 或 zip 描述。
+
+不要在消息里输出文件的完整内容、工作区保存路径或裸链接。