← 返回 Skills 市场
kedoupi

Insight Radar (洞察雷达)

作者 KeDouPi(珂抖屁) · GitHub ↗ · v1.0.6 · MIT-0
cross-platform ✓ 安全检测通过
185
总下载
0
收藏
0
当前安装
7
版本数
在 OpenClaw 中安装
/install insight-radar
功能描述
Provides AI-driven strategic extraction of news concepts and patterns to evolve AI thinking, plus CORE-analyzed personalized news briefings by category.
使用说明 (SKILL.md)

Insight Radar (洞察雷达)

Two modes, one skill:

🧠 For AI Self-Evolution

扫描新闻,提取概念与战略模式,写入知识库:

  • Concepts (主要) → concepts.md: 当前事件、技术发展、市场变化
  • Thinking Patterns (偶尔) → patterns/*.md: 仅当新闻揭示可复用模式时

📰 For User Learning

CORE 分析的战略简报,不是新闻聚合:

  • [C] Core Logic: 第一性原理(不是发生了什么,而是 WHY)
  • [O] Opportunity: 价值流向,谁赢了?
  • [R] Risk: 反共识,隐藏黑天鹅
  • [E] Execution: 你今天该做什么?

News categories: 由 news-source-manager 管理,默认 AI/Tech。


Workflow

1. Load News Config

调用 news-source-manager 获取用户的活跃新闻类别:

{
  "active_categories": [{
    "name": "AI/Tech",
    "keywords": ["AI", "machine learning", "LLM", "AI agents", "semiconductor", "robotics"],
    "sources": ["TechCrunch", "MIT Tech Review"],
    "search_params": {"count": 5}
  }]
}

If news-sources.json 不存在: 触发 news-source-manager 初始化流程。 Fallback: 用户跳过配置时,默认 "AI/Tech" 类别。


2. Search Recent News

搜索策略: 宽泛扫描 + 智能补搜 + 中英双语。

核心原则: 不预定义固定 hints。越简单的 query 返回越全面的结果(测试验证:"AI technology news March 2026" 比精心设计的关键词组合覆盖度更高)。

For each active category, 分两步:

Step 2a: 宽泛扫描(每类别固定 2 次)

英文: "{category_name} news {current_month} {current_year}"
中文: "{类别中文名} 新闻 {current_year}"
count: 5

示例 (Finance/Crypto, 当前 March 2026):
  WebSearch("Finance Crypto news March 2026", count: 5)
  WebSearch("金融 加密货币 新闻 2026", count: 5)

类别名称映射:

category_name 类别中文名
AI/Tech AI 科技
Business Strategy 商业战略
Finance/Crypto 金融 加密货币
Health/Bio 生物医药 数字健康
Energy/Climate 新能源 气候科技
Policy/Regulation AI监管 科技政策
Product Design 产品设计 用户体验

Step 2b: 覆盖度评估 + 智能补搜(每类别 0-2 次)

扫描 Step 2a 结果后,用 news-sources.json 中的 keywords 评估覆盖度:

1. 将该类别的 keywords 分为 3-4 个子领域簇
   例如 AI/Tech: [模型/Agent] [芯片/硬件] [安全/治理] [应用/商业]

2. 检查 Step 2a 结果覆盖了哪些簇
   例如: 结果中有 GPT-5.4、Agent 部署 → [模型/Agent] ✅
         结果中没有芯片相关 → [芯片/硬件] ❌

3. 对未覆盖的簇,生成 1 条补搜 query
   例如: WebSearch("AI chips semiconductor news March 2026", count: 3)

4. 最多补搜 2 次/类别(硬上限,防止无限膨胀)

总调用量: 3 类别 × 2 底线 + 0-6 补搜 = 6-12 次(实际通常 8-9 次)。

Fallback

仅当某类别 Step 2a + 2b 合计 \x3C 2 条结果时,将 {current_month} 改为上月重搜。

If ALL categories fail: 通知用户 "过去 24h 无重大新闻,建议扩展类别或时间范围。"

后处理

去重: 合并所有类别结果,按 URL/title 去重,优先保留高权威来源(sources 中 priority: 1)。

URL 验证: 检查返回 URL 是否为具体文章。首页/分类页 → 用 web_fetch 提取具体链接。

战略价值筛选 — 从去重结果中选 TOP 3-5 条:

高价值信号 (优先): 揭示行业本质变化 / 反共识观点 / 跨领域影响 / 基础设施机会 低价值信号 (跳过): 纯融资公告 / 常规产品发布 / 无新洞察的共识观点

类别覆盖: 选出的 3-5 条至少覆盖 2 个不同类别。如果都来自同一类别 → 补 1 条次要类别。

输出格式: 每条标注类别、时间、来源链接

### 1️⃣ 中国AI Agent爆火 【AI/Tech】
📅 1 day ago (2026-03-22)
🔗 [Reuters](https://www.reuters.com/...)

3. Analyze with CORE Framework

对每条选中的新闻,调用 core-prism skill 进行 CORE 分析。传入 USER.md 中的用户上下文用于 [E] Execution 维度个性化。


4. Generate Strategic Briefing

4 个 mandatory sections:

  1. 📰 核心资讯 (Core News)

    • 3-5 items with CORE analysis
    • 每条含: 类别标签【】, 时间 📅, 来源链接 🔗, C/O/R/E 分析
  2. 🎯 战略简报 (Strategic Briefing)

    • 今日破局点: 这些故事一起揭示了什么?
    • So What? 给用户一个尖锐的可操作问题
  3. 🧠 认知沉淀 (Cognitive Digest)

    • 今日关键事件摘要 (写入 news-log)
    • 候选概念 (如果今日新闻揭示了一个全新的认知框架,标记为候选,累积验证后才入 concepts.md)
    • 候选模式 (如果同一模式已在 3+ 条不同新闻中反复出现,才提炼入 patterns/*.md)
    • 判断力复盘 (哪些旧假设被验证/推翻?)
  4. 🤔 盲区质询 (Blind Spot Questions)

    • 3-5 个挑战用户假设的思考题

5. Write to Knowledge Base — 三层架构

每日新闻和深度认知分层存储,避免知识库被新闻复述淹没。

三层写入规则

层级 文件 入库标准 频率
Layer 1: 新闻日志 memory/news-log/YYYY-MM-DD.md 每条分析过的新闻 每天
Layer 2: 概念 memory/knowledge-base/concepts.md 真正的新认知框架,不是事件记录 每周 1-3 个
Layer 3: 思维模式 memory/knowledge-base/patterns/*.md 同一模式在 3+ 条不同新闻中反复出现 每月 1-2 个

Step 5.1: 写入新闻日志(每次必须)

创建/追加 ~/.openclaw/workspace/memory/news-log/YYYY-MM-DD.md:

# 资讯早报 - YYYY-MM-DD

## 1. [新闻标题] 【类别】
- **来源**: [链接]
- **CORE 摘要**: C: ... | O: ... | R: ... | E: ...
- **候选概念**: [如有,标记为候选] 或 "无"
- **关联旧模式**: [如能关联到 patterns/*.md 中已有模式] 或 "无"

Step 5.2: 概念入库(高门槛,非每日)

入库条件(必须同时满足):

  • ✅ 这是一个新的认知框架,不是事件/产品/公司名
  • ✅ 具有持久解释力(6 个月后仍然有用)
  • ✅ 无法用已有概念解释

Bad: "LightGen 光芯片"(这是一条新闻) Good: "光计算范式"(这是一个持久概念,但需要更多数据点确认才入库)

暂不入库的标记为"候选概念"记录在 news-log 中,等累积验证。

Step 5.3: 思维模式入库(最高门槛)

入库条件: 同一模式在 3+ 条不同日期的新闻中反复出现

在 news-log 中标记 "关联旧模式",当累积到 3+ 条时,提炼入 patterns/{id}.md

写入格式必须与 cognitive-forge 一致(YAML frontmatter + 7 字段正文):

---
id: {kebab-case-english}
name_zh: {中文名}
name_en: {English Name}
source: Insight Radar, {触发该模式的 3+ 条新闻来源概述}
category: {investing|startup|systems|ai-thinking|positioning|management|growth|cognitive-bias|influence|economics}
tags: [{3-5个中文标签}]
scenarios: [{3-5个应用场景}]
related_models: [{2-4个已有模型id}]
difficulty: {beginner|intermediate|advanced}
date: {YYYY-MM-DD}
---

**核心逻辑**: {模式的本质,一段话}

**思维框架**: {核心机制,一句话}

**决策原则**: {在XX场景下,应该XX而非XX}

**盲区警告**: {何时失效}

**反射弧**: {看到XX信号 → 联想到模型 → 行动}

**锚定案例**: {触发该模式的真实新闻案例}

**反共识**: {❌ "旧常识" → ✅ 新真相}

注意:insight-radar 产出的模式 source 字段标注为 "Insight Radar, ..."(而非书籍来源),以区分来源渠道。

Step 5.4: 验证写入

写入后,用 Read 工具读取 news-log 文件确认条目存在。

Completion Checkpoint:

  • □ news-log/YYYY-MM-DD.md 已写入?
  • □ 概念入库?(仅当满足高门槛条件时)
  • □ 思维模式入库?(仅当同一模式 3+ 次出现时)
  • □ 用 Read 验证了写入?

6. Deliver Briefing

通过触发通道发送给用户。

用户上下文个性化 (强制步骤):

  1. 读取 USER.md (~/.openclaw/workspace/USER.md):

    • 工作经历/现在 → profession
    • 兴趣/爱好 → interests
    • 当前焦虑/未来规划 → challenges
    • 不存在 → 使用通用第二人称
  2. 映射到 [E] Execution:

    • 引用职业: "如果你在做 AI 产品..."
    • 关联兴趣: "鉴于你对 fintech 的关注..."
    • 回应挑战: "对于你建立第二曲线的目标..."
  3. 始终使用第二人称 ("你/你的")


Configuration

优先级 来源 用途
1 memory/news-sources.json 新闻类别 (由 news-source-manager 管理)
2 USER.md 个性化 [E] Execution 维度
3 默认值 AI/Tech 类别, 通用第二人称

Knowledge base paths (auto-created if missing):

  • ~/.openclaw/workspace/memory/knowledge-base/concepts.md
  • ~/.openclaw/workspace/memory/knowledge-base/patterns/*.md

Quality Standards

Forbidden: 新闻聚合无分析 / 表面评论 / 重复新闻稿 / "AI is changing everything" 式空话 / 空 briefing

Required: 一针见血的洞察 / 反共识视角 / 可操作 / 跨文章找暗线 / 最少 2-3 条新闻 / 🧠 认知库更新 section is MANDATORY


Execution Flow

Daily trigger
    ↓
1. Load news-source-manager config
    ↓
2. Search recent news (search_hints + fallback)
    ↓
3. Call core-prism for CORE analysis
    ↓
4. Generate briefing (4 mandatory sections)
    ↓
5. Write to knowledge base (MANDATORY)
    ├─ concepts.md
    ├─ patterns/*.md (if patterns found)
    └─ Verify with Read
    ↓
6. Deliver briefing (personalized via USER.md)

References

  • See example-output.md for output format
  • See category-config.md for adding custom categories
  • See core-prism skill for detailed CORE framework
  • See news-source-manager skill for category management
安全使用建议
This skill appears internally consistent and focuses on local news scanning and knowledge-base updates. Before enabling it, check these practical points: (1) USER.md is used to personalize 'Execution' — remove any secrets/PII from that file if you don't want them written into logs; (2) the skill will create/append files under ~/.openclaw/workspace/memory/ (news-log, knowledge-base) — ensure you are comfortable with those files existing on disk and being readable by the agent; (3) it calls two other skills (core-prism, news-source-manager) — install or review those dependencies first so you trust their behavior; (4) it uses the agent's built-in WebSearch/WebFetch to pull article content — verify that no external API keys or unknown endpoints are configured in your environment; and (5) if you later integrate delivery (HEARTBEAT-news.md / Feishu), review that integration separately to avoid unintended external sharing.
功能分析
Type: OpenClaw Skill Name: insight-radar Version: 1.0.6 The insight-radar skill is a news intelligence system designed to aggregate news, perform structured analysis (CORE framework), and maintain a local knowledge base. While it requests permissions to read user context (USER.md) and access the network (WebSearch/WebFetch), these are strictly aligned with its stated purpose of providing personalized briefings and fetching current events. The 'AI self-evolution' logic is a standard local data-persistence pattern that writes structured insights to markdown files within the workspace (SKILL.md), and there is no evidence of data exfiltration, malicious execution, or harmful prompt injection.
能力评估
Purpose & Capability
Name/description match the declared behavior: searches news, runs CORE analysis via core-prism, and writes daily logs and high-bar knowledge entries. Declared dependencies (core-prism, news-source-manager) and the memory files it reads/writes are coherent with the stated purpose.
Instruction Scope
SKILL.md explicitly limits actions to: use built-in WebSearch/WebFetch for news, call news-source-manager and core-prism, read USER.md and news-sources.json for personalization/config, and write files under ~/.openclaw/workspace/memory (news-log, concepts, patterns). There are no instructions to read unrelated system files, request credentials, or exfiltrate data to hidden endpoints.
Install Mechanism
Instruction-only skill with no install spec and no downloaded code. Low install risk — nothing is written to disk beyond the skill's normal memory writes documented in SKILL.md.
Credentials
No environment variables, no credentials, and no config paths beyond workspace memory files are requested. Use of USER.md for personalization is expected but worth noting for privacy.
Persistence & Privilege
Does not request always:true or elevated platform privileges. It writes only to its own workspace memory paths as documented and relies on other skills for analysis; autonomous invocation is default but not combined with additional risky privileges here.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install insight-radar
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /insight-radar 触发
  4. 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
v1.0.6
**insight-radar 1.0.6 changelog** - Added explicit network permissions for WebSearch and WebFetch, enabling direct news searching and article content validation. - Specified that no external API keys/credentials are needed; all search functions use agent's built-in tools. - Clarified that external_outputs are not used—news briefings are generated locally, with delivery handled externally (e.g., by HEARTBEAT-news.md). - Updated documentation to reflect stricter permissions structure, enhancing transparency and security. - No changes to workflow, user experience, or core functionality.
v1.0.5
- Strategic pattern storage updated: thinking patterns now saved in individual files under patterns/*.md instead of a single thinking-patterns.md file. - Pattern files must use the cognitive-forge YAML frontmatter + 7-field format, with “source” attributed to “Insight Radar”. - All references and workflow steps aligned to new patterns/*.md structure. - No logic or config changes beyond knowledge base pattern file management.
v1.0.4
- Removed external database (Feishu Bitable) integration steps and related configuration. - Updated the permissions and configuration sections to reflect the removal of ENV vars related to external services. - The workflow and execution flow now focus solely on local logging and knowledge base updates. - Documentation and configuration priority tables updated to match local-only architecture.
v1.0.3
Version 1.0.3 - Added a dedicated config section for explicit read/write resource configuration and environment variables. - Clarified and annotated permissions to show purpose (e.g., which files are for user context or knowledge base layers). - No changes to core logic, workflow, or external interface. - No file-level/in-code changes detected; SKILL.md now better documents config and operational permissions.
v1.0.2
**insight-radar v1.0.2 Changelog** - Added explicit file read/write and environment variable permission structure in SKILL.md. - Clarified and optimized the news search and supplementation workflow: now requires both broad and targeted queries (including bilingual, multi-subdomain coverage, and smart fallback logic). - Fully restructured the tiered knowledge base write rules—daily news, concepts (high threshold), and thinking patterns (highest threshold)—with detailed, non-redundant writeback requirements. - Enhanced the briefing output requirements with four mandatory, deeply analyzed sections: Core News, Strategic Briefing, Cognitive Digest, and Blind Spot Questions. - Strengthened rules for category diversity, strategic value filtering, repeat-detection for pattern mining, deduplication, and personalization based on USER.md. - Improved documentation of default/fallback logic, configuration hierarchy, and execution flow, with precise mandatory/forbidden standards for output quality.
v1.0.1
insight-radar v1.0.1 - Added dependency declarations for `core-prism` and `news-source-manager`. - Defined explicit read/write config for user data and knowledge base outputs. - Introduced support for environment variables (`FEISHU_APP_TOKEN`, `NOTION_API_KEY`) for outbound integration. - No logic or feature changes; this update standardizes configuration and dependency documentation.
v1.0.0
insight-radar 1.0.0 – Initial Release - Launches a dual-purpose news intelligence system: supports both AI self-evolution and highly customized user news briefings. - Introduces CORE analysis framework (Core Logic, Opportunity, Risk, Execution) for strategic news summaries tailored to user interests. - Captures and stores concepts and strategic patterns from news for continuous AI learning and future application. - Enables full customization of news categories, sources, and frequency via user config file or environment variables. - Robust news sourcing pipeline ensures only high-value, recent, diverse news is surfaced; includes both fallback logic and quality filtering. - Clear workflow and output format designed for readability, strategic value, and category diversity.
元数据
Slug insight-radar
版本 1.0.6
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 7
常见问题

Insight Radar (洞察雷达) 是什么?

Provides AI-driven strategic extraction of news concepts and patterns to evolve AI thinking, plus CORE-analyzed personalized news briefings by category. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 185 次。

如何安装 Insight Radar (洞察雷达)?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install insight-radar」即可一键安装,无需额外配置。

Insight Radar (洞察雷达) 是免费的吗?

是的,Insight Radar (洞察雷达) 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Insight Radar (洞察雷达) 支持哪些平台?

Insight Radar (洞察雷达) 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 Insight Radar (洞察雷达)?

由 KeDouPi(珂抖屁)(@kedoupi)开发并维护,当前版本 v1.0.6。

💬 留言讨论