Info Magnet

Info Magnet — set up topics you care about and let information come to you. Supports web search, RSS feeds, and URL monitoring. Periodic scans push relevant...

Security Scan

Info Magnet's declared purpose (periodic web/RSS/URL monitoring and summarization) matches its instructions and included Python script; it requests no credentials and uses only local storage and normal outbound HTTP requests.

Purpose & Capability

Name/description (monitoring web/RSS/URLs, periodic scans, digests) align with the included SKILL.md and the Python script. Required binaries are only python and no credentials or unrelated binaries are requested.

Instruction Scope

Runtime instructions are scoped to creating/listing/managing magnets, fetching RSS, simple webpage checks, and producing digests. The agent is expected to perform web_search for web-result enrichment; this is explicitly documented. The skill stores state under ~/.openclaw/memory which is consistent with its purpose. It does not instruct the agent to read unrelated system files or secrets.

Install Mechanism

No install spec (instruction-only with a shipped Python script). No downloads or external installers are invoked. This is the lowest-risk install pattern for this functionality.

Credentials

No environment variables, API keys, or credentials are requested. The skill only needs outbound HTTP access to fetch RSS/web pages and relies on agent-provided web_search for searches; that is proportionate to the stated capabilities.

Persistence & Privilege

always: false and model-invocation is allowed (platform default). The skill writes its own data under ~/.openclaw/memory (magnets.json, magnet-seen.jsonl, magnet-digests/) which is expected. It does not attempt to modify other skills or system-wide agent settings.

Assessment

This skill appears to do what it says: it will make outbound HTTP requests to RSS feeds, URLs you configure, and expects the agent to run web_search for web lookups. Before enabling: (1) review the full infomagnet.py (it’s included) if you need to confirm exact behavior; (2) avoid adding magnets that point to private/internal URLs or to feeds containing sensitive tokens, since the skill will fetch those addresses; (3) if you run agents in sensitive environments, consider restricting network access or running the skill in a sandbox; (4) integrating the skill with your HEARTBEAT or message channel will cause it to run periodically — ensure you’re comfortable with that cadence and stored summaries under ~/.openclaw/memory.

信息磁铁 / Info Magnet

不是你去找信息，是信息来找你。

Don't search for information. Set up magnets that attract it.

核心理念

传统方式：每天手动搜一遍 → 重复搜同样的东西 → 漏了重要更新
磁铁方式：设好关注点 → 自动定期扫描 → 有新的才推送 → 没有就安静

命令参考

MAGNET = python <skill_dir>/scripts/infomagnet.py

创建磁铁

# Web 搜索型：定期搜索关键词
$MAGNET add --name "AI 动态" --topics "大模型,AI agent,GPT-5" --sources web --schedule daily

# RSS 订阅型：监控 RSS 源
$MAGNET add --name "Hacker News 热帖" \
  --topics "tech" \
  --sources rss \
  --rss "https://hnrss.org/frontpage" \
  --schedule hourly

# URL 监控型：检测网页更新
$MAGNET add --name "OpenClaw 更新" \
  --topics "openclaw" \
  --sources url \
  --urls "https://github.com/openclaw/openclaw/releases" \
  --schedule daily

参数说明：

• --name：磁铁名称
• --topics：关注的关键词（逗号分隔）
• --sources：信源类型（web/rss/url，逗号分隔）
• --schedule：检查频率（hourly/daily/weekly）
• --rss：RSS feed URL
• --urls：要监控的网页 URL
• --notes：备注

查看所有磁铁

$MAGNET list

扫描新内容

# 扫描所有活跃磁铁
$MAGNET scan --all

# 只扫描指定磁铁
$MAGNET scan --id mag_xxx

扫描结果会自动保存为摘要文件，已看过的 URL 不会重复出现。

查看摘要

# 查看最近 24 小时的内容
$MAGNET digest

# 查看最近 7 天
$MAGNET digest --since 7d

# 只看某个磁铁
$MAGNET digest --id mag_xxx --since 48h

管理磁铁

# 暂停
$MAGNET pause --id mag_xxx

# 恢复
$MAGNET resume --id mag_xxx

# 删除
$MAGNET remove --id mag_xxx

# 标记已读（手动标记某个 URL）
$MAGNET mark-read --url "https://..."

导入 RSS

$MAGNET import-rss --url "https://rss.example.com/feed.xml" --name "我的订阅"

Agent 集成工作流

定期扫描（Heartbeat 集成）

在 HEARTBEAT.md 中添加：

## 信息磁铁扫描
- 运行 `python ~/.openclaw/skills/info-magnet/scripts/infomagnet.py scan --all`
- 如果有新发现，发送摘要给用户
- 如果没有，跳过

扫描 + 搜索的完整流程

1. $MAGNET scan --all
   → 生成待搜索的 topic 列表（web-pending 条目）

2. Agent 对每个 web-pending 执行 web_search
   → 过滤掉已见过的 URL
   → 有新的就记录

3. $MAGNET digest --since 24h
   → 汇总所有新发现

4. 推送给用户：
   "📬 今日信息摘要：发现 3 条你可能感兴趣的内容..."

智能过滤

Agent 在执行搜索后，不是把所有结果都推给用户，而是：

1. 去重：跳过已见过的 URL
2. 相关性判断：根据用户的关注点筛选
3. 摘要：每条内容用 1-2 句话概括
4. 分类：按磁铁分组展示

示例配置

# 技术资讯
$MAGNET add --name "技术前沿" \
  --topics "rust,webassembly,ai agent,llm" \
  --sources web,rss \
  --rss "https://hnrss.org/best,https://lobste.rs/rss" \
  --schedule daily

# 项目监控
$MAGNET add --name "我的依赖库更新" \
  --topics "playwright update,flask release" \
  --sources web \
  --schedule weekly

# 行业动态
$MAGNET add --name "AI 产品动态" \
  --topics "new ai product,chatgpt plugin,claude api" \
  --sources web \
  --schedule daily

# 中文社区
$MAGNET add --name "掘金热帖" \
  --topics "前端" \
  --sources rss \
  --rss "https://rsshub.app/juejin/category/frontend" \
  --schedule daily

数据存储

~/.openclaw/memory/
├── magnets.json           # 磁铁配置
├── magnet-seen.jsonl      # 已见过的 URL（去重用）
└── magnet-digests/        # 历史摘要
    ├── mag_xxx_1775095389.json
    └── ...

设计哲学

• 推 > 拉：信息主动推送，而不是你去找
• 安静：没有新内容就不打扰
• 去重：同一内容只出现一次
• 可组合：支持 web 搜索 + RSS + URL 监控混搭
• 零依赖：纯 Python 标准库

局限性

• Web 搜索依赖 Agent 的 web_search 工具（需要在 Agent 环境中执行）
• RSS 解析不支持所有 feed 格式（兼容 RSS 2.0 和 Atom）
• URL 监控是简单 hash 比较，不支持动态页面
• 没有内置推送渠道，需要 Agent 通过消息通道转发