idea-inbox

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。

Security Scan

The skill's code and runtime instructions match its stated purpose (collect Feishu DM ideas, classify with an LLM, and write to a Bitable); nothing requested or installed is disproportionate, though it will read local model/config files and send idea text to your configured model provider and persist Bitable config locally.

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.

Assessment

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.

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。

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

1. 校验触发条件（私聊 + 前缀匹配）
2. 抽取原文内容（去掉前缀）
3. 调用大模型生成严格 JSON：{category,tags,summary}
4. 标签自动新增：若 tags 不在 Bitable 选项中，先追加选项
5. 写入 Bitable：内容/AI归纳/类别/标签/状态=收集箱/来源=飞书私聊
6. 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