对话生产器（自用）

从零生成模拟微信群聊记录的 Excel (.xlsx) 文件，包含 group_info / active_members / message_stream 三个 sheet，格式与派平台的群聊训练数据完全兼容。使用场景：(1) 需要生成 AI 助手训练数据的模拟群聊对话；(2) 测试群聊 FAQ/知识库系统；(...

Security Scan

The skill's code, runtime instructions, and dependencies align with its stated purpose (generating simulated WeChat group chat .xlsx files); it runs locally, asks for no credentials, and contains no network endpoints or exfiltration behavior, though there are minor implementation oddities to be aware of.

Purpose & Capability

Name/description promise (generate group chat Excel compatible with target platform) matches included files: generator and xlsx-writing scripts plus format docs. Requested runtime (Node + xlsx) is appropriate for the stated task.

Instruction Scope

SKILL.md stays within the task: define group, spawn subagents to create messages, then write an xlsx. The only notable runtime instruction outside pure generation is guidance to use sessions_spawn(runtime="subagent"...), which is a platform-specific multi-agent mechanism (expected for multi-role generation). No instructions ask the agent to read unrelated system files, environment variables, or contact external endpoints.

Install Mechanism

There is no install spec in the registry (instruction-only), but the code expects the npm package 'xlsx' and documents an npm install step into /tmp/xlsxparse. This is a reasonable, low-risk approach, but the code hardcodes require('/tmp/xlsxparse/node_modules/xlsx') as the default module path which is unusual and brittle; the script allows overriding via a CLI arg.

Credentials

Skill requests no environment variables, no credentials, and no config paths. The only filesystem interactions are reading/writing local JSON and XLSX files (paths provided via CLI or defaults). This is proportionate to the functionality.

Persistence & Privilege

Skill is not always-enabled and uses normal autonomous invocation settings. It does not modify other skills or system-wide config. It writes output files locally (expected behavior) and suggests optionally installing a local npm dependency in /tmp, which does not imply persistent privileged presence beyond that.

Assessment

This skill appears to do exactly what it says: locally generate simulated group chat .xlsx files. Before installing or running: 1) Be prepared to run Node.js and npm; the scripts expect the 'xlsx' package (the docs recommend installing it under /tmp/xlsxparse). Installing that package will download code from the npm registry — normal for Node tooling but review if your environment restricts external downloads. 2) Note the scripts are local and read/write files you specify (group_def.json, messages.json, output.xlsx); avoid running them in directories with sensitive files or letting them overwrite important files. 3) The code hardcodes a default module path (/tmp/xlsxparse/node_modules/xlsx) and there is a minor CLI bug in generate_group_def.js argument handling — nothing malicious, but you may want to inspect/adjust the scripts (or pass explicit --group/--messages/--output/--xlsx args) before use. 4) No network endpoints, secrets, or exfiltration behavior were found, and no environment variables are required. If you need higher assurance, run the scripts in an isolated environment (container) and review the installed npm package contents before execution.

Chat Record Generator

生成符合派平台格式的模拟群聊 Excel 文件，支持多角色 multi-Agent 并行生成对话内容。

快速流程

1. 定义群组 → 2. 并行生成对话（multi-Agent） → 3. 编排 + 写入 xlsx

---

第一步：定义群组

需要确认以下信息（如用户未提供，按默认值生成）：

字段	说明	示例
群名	任意字符串	AI资讯交流群
群ID	数字字符串	123456789
成员数	建议 5–15 人	10
AI助手昵称	群内机器人	元宝
话题列表	2–4 个话题	见下方
目标消息总数	建议 50–200 条	150

运行 scripts/generate_group_def.js 生成 group_definition.js，或参考 references/group-schema.md 手动编写。

---

第二步：并行生成对话（multi-Agent）

最大并发限制：5 个 subagent，超过时分批启动。

用 sessions_spawn(runtime="subagent", mode="run") 为每个角色启动一个子 Agent。

Prompt 模板

你是「{群名}」中的角色【{昵称}】，{年龄}岁{职业}。

语言风格：{风格描述}

【群聊话题】
话题1（第1-N轮）：{话题1标题}
话题2（第N+1-M轮）：{话题2标题}
...

【你的任务】
以{昵称}身份生成{N}条消息，每个话题约{N/话题数}条。
- 话题1：{角色在该话题的具体行为}
- 话题2：...

直接输出JSON数组，不要有其他文字：
[{"话题": 1, "发言人": "{昵称}", "内容": "消息内容"}, ...]
共{N}条。

消息分配参考（总140条，10角色）

角色类型	条数
活跃成员（核心讨论者）	15
普通成员	12–14
工具人（发起话题/@AI）	15

---

第三步：编排 + 写入 xlsx

收集所有子 Agent 结果后，运行 scripts/write_xlsx.js 写入文件。

关键规则

• 消息类型：text → 内容非空、附件 null；image/file → 内容 null、附件为 image_N/file_N
• 时间分配：每条消息间隔约 0.002（约3分钟），话题间隔 0.05–0.1（约1–2小时）
• ⚠️ numFmt Bug：必须对每个时间单元格手动设置 z 属性，aoa_to_sheet 不会保留格式
• Excel 时间 epoch：new Date(Date.UTC(1899, 11, 30))

时间格式

// 时间列
cell.z = 'yyyy/m/d h:mm;@'
// join_time 列（含秒）
cell.z = 'yyyy/m/d h:mm:ss;@'

---

xlsx 格式规范

参见 references/xlsx-format.md 获取完整的三个 sheet 字段定义。

---

依赖

• Node.js + xlsx 库（npm install xlsx，通常缓存在 /tmp/xlsxparse/node_modules/xlsx）
• 若未安装：mkdir -p /tmp/xlsxparse && cd /tmp/xlsxparse && npm install xlsx

---

注意事项

• 昵称风格：使用真实微信群昵称（陈默、林晓、周宇老师），避免社交媒体 ID 风格
• @元宝：AI助手被 @时应在同一话题内接近的轮次生成回复消息（可由主 Agent 补写）
• task_list：始终为 null，但列必须保留
• bot_config：group_info sheet 中为 null