← 返回 Skills 市场
idea-inbox
作者
xiaoxiaoxi
· GitHub ↗
· v1.1.1
· MIT-0
132
总下载
1
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install idea-inbox
功能描述
收集“idea:/灵感:”消息到飞书多维表格(默认自动创建新表),用大模型生成AI归纳/类别/标签(支持自动新增标签),并按配置的每日时间(默认10:02,今日新增=0不发)推送当日汇总。
使用说明 (SKILL.md)
灵感妙计(Idea Inbox)
把你在飞书私聊里随手发来的想法收集成结构化的灵感池,并自动分类,方便后续筛选与推进。
Triggers
- 以
idea:开头的消息 - 以
灵感:开头的消息
仅在飞书私聊生效。
Use Cases
- 你随手发一个想法:我自动入库并分类
- 你持续积累一周:我每天 10:02 给你发当日新增汇总(若当日 0 新增则不发)
Data Store
- 默认:首次运行(第一次发
idea:/灵感:)自动创建一个新的飞书多维表格(Bitable)App:灵感妙计。 - 创建完成后,会把
app_token/table_id/字段id等写入本地配置文件:~/.openclaw/idea-inbox/config.json
字段(默认):内容 / AI归纳 / 类别 / 标签 / 状态 / 来源 / 创建时间
Configuration
~/.openclaw/idea-inbox/config.json(自动生成,可手改)示例:
{
"prefixes": ["idea:", "灵感:"],
"daily_time": "10:02",
"bitable": {
"app_token": "...",
"table_id": "...",
"fields": {
"content": "内容",
"ai_summary": "AI归纳",
"category": "类别",
"tags": "标签",
"status": "状态",
"source": "来源",
"created_time": "创建时间"
}
},
"model": {
"mode": "codex_files_first",
"fallback": null
},
"tags": {
"auto_add": true,
"max_tags": 5
},
"status": {
"options": ["待处理", "已处理", "无法处理"],
"default": "待处理"
}
}
daily_time:仅需配置每日时间;send_if_zero固定为 false(0 新增不发)。- 模型:默认优先读取
~/.codex/auth.json+~/.codex/config.toml(不读 env)。- 若用户没有 codex 文件,可在
model.fallback填baseUrl/apiKey/apiType/model。
- 若用户没有 codex 文件,可在
Inputs
- Required:
text: 原始消息文本(包含前缀)chat_type: 需为 direct/p2p(私聊)
Outputs
- 成功:写入一条记录到 Bitable,并返回 record_id
- 失败:返回错误原因(网络/权限/字段缺失/模型调用失败等)
Workflow
- 校验触发条件(私聊 + 前缀匹配)
- 抽取原文内容(去掉前缀)
- 调用大模型生成严格 JSON:{category,tags,summary}
- 标签自动新增:若 tags 不在 Bitable 选项中,先追加选项
- 写入 Bitable:内容/AI归纳/类别/标签/状态=收集箱/来源=飞书私聊
- 每日 10:02 汇总:统计当日新增,>0 才推送
Limitations
- 本技能不处理群聊消息
- 标签自动新增会带来标签膨胀;可后续加“标签黑名单/合并”规则
- 模型输出需严格 JSON;会做枚举校验与截断
安全使用建议
This skill appears internally consistent and does what it says. Important things to consider before installing:
- The skill will read local model/provider config files (~/.codex/config.toml, ~/.codex/auth.json and your OpenClaw config) and use any API keys it finds to call the model endpoint; it will send the idea text to that model provider. Make sure you trust the model provider configured in those files.
- On first run it will create a Feishu (Lark) Bitable app and save app_token/table_id and field ids to ~/.openclaw/idea-inbox/config.json. That file may contain sensitive tokens — review and restrict file permissions if needed.
- If you do not want idea text sent to an LLM, you can disable AI in the config (ai.enabled=false) so the skill falls back to local rules.
- Review ~/.codex/auth.json and OpenClaw provider entries to confirm where data will be sent. If you have confidential ideas, ensure the configured model provider is appropriate or disable AI classification.
Confidence is high for this assessment because the code, SKILL.md, and file manifest are coherent and no installer or unexpected external URLs are present. If you want a lower-privilege mode or proof of exactly which provider will be used at runtime, provide a sample ~/.codex/config.toml and ~/.openclaw/openclaw.json (or confirm they are absent) so we can re-evaluate based on real provider entries.
能力评估
Purpose & Capability
Name/description match code and SKILL.md: triggers on DM prefixes, classifies items (rules or LLM), creates/updates a Feishu Bitable, and writes local config. No unrelated credentials or binaries are required.
Instruction Scope
SKILL.md and scripts explicitly read and write local configs (e.g., ~/.openclaw/idea-inbox/config.json, ~/.codex/config.toml, ~/.codex/auth.json, and OpenClaw config). The skill sends idea text to the configured model provider (base_url + apiKey) for LLM classification. These behaviors are consistent with the stated purpose but are material privacy/actions (sending content to an external model and persisting tokens/configs).
Install Mechanism
Instruction-only skill (no install spec). Code files are present but there is no remote download or install step. Nothing is fetched from arbitrary URLs during setup.
Credentials
The skill declares no required env vars, but the code intentionally reads local config files (~/.codex/* and OpenClaw config) and allows env overrides (OPENCLAW_CONFIG, CODEX_CONFIG, CODEX_AUTH, IDEA_INBOX_*). This is proportionate to needing model/provider credentials, but users should note the skill will use any model API keys present in those files and may write Bitable tokens into ~/.openclaw/idea-inbox/config.json.
Persistence & Privilege
always:false and it only writes its own config under the user's home (~/.openclaw/idea-inbox). It does not request to modify other skills or global agent settings.
如何使用
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install idea-inbox - 安装完成后,直接呼叫该 Skill 的名称或使用
/idea-inbox触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.1.1
No changes detected in this version.
- No file changes or updates in documentation.
- Functionality and description remain the same as previous version.
v1.1.0
v1.1.0: Feishu idea inbox skill. Auto-creates a new Bitable on first run, generates AI summary/category/tags with rules fallback, supports auto-adding tags, and daily digest time config (0 new items -> no message).
元数据
常见问题
idea-inbox 是什么?
收集“idea:/灵感:”消息到飞书多维表格(默认自动创建新表),用大模型生成AI归纳/类别/标签(支持自动新增标签),并按配置的每日时间(默认10:02,今日新增=0不发)推送当日汇总。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 132 次。
如何安装 idea-inbox?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install idea-inbox」即可一键安装,无需额外配置。
idea-inbox 是免费的吗?
是的,idea-inbox 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
idea-inbox 支持哪些平台?
idea-inbox 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 idea-inbox?
由 xiaoxiaoxi(@hanxiaolin)开发并维护,当前版本 v1.1.1。
推荐 Skills