SkillFoxx

FactReach skill

A skill and CLI for agents: search and reading across 23 channels, from web and GitHub to social networks, with an evidence rule.

  • 60 ★ on the repository
  • Last commit
  • MIT
Skill

Broad access

Why this level

  • It reads social and gated pages through your session and cookies, so a platform may restrict the account.
  • System installation changes the environment and writes rules into `CLAUDE.md` and `AGENTS.md`.
  • Queries go to outside search and page reading services.
All reasons and checks

simonlin1212/factreachAuthor:Simon Lin

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add skills/factreach

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • It reads social and gated pages through your session and cookies, so a platform may restrict the account.
  • System installation changes the environment and writes rules into `CLAUDE.md` and `AGENTS.md`.
  • Queries go to outside search and page reading services.

This is third-party code. Review the repository files before installing.

What it does

FactReach gives an agent access to outside facts through 23 channels: web search, ordinary pages, GitHub, Hacker News, Stack Overflow, papers, RSS, YouTube, Reddit, X, Weibo, Xiaohongshu, Douyin and more. Keyless search falls back between DuckDuckGo, Brave and Bing, pages are read through a Jina Reader, Firecrawl and browser-harness chain, and the real Chrome is enabled only explicitly. `factreach doctor --json` reports each channel state with a repair hint. A separate rules layer writes a "search first, then answer" block into `CLAUDE.md` or `AGENTS.md`. The policy commands show the change first and write only after your consent.

Who it is for. People building a research agent who want it to look facts up on the web and Chinese social networks instead of answering from model memory.

Good fit when

  • An agent must check fresh data and answer with links instead of from training data.
  • You need to read a Xiaohongshu, Douyin or Bilibili post or video through routing over your own session.
  • A page is dynamic and a plain fetch cannot get it.
  • After installation you want one command showing which channels work and what to repair.

Not a fit when

  • Queries must not go to outside search services.
  • You need guaranteed social network access: platforms can limit automated requests.
  • The agent needs neither social networks nor dozens of channels and a single search MCP would do.

Example request

Find what Hacker News, Reddit and Xiaohongshu say about this library release and give the links.

Limitations

Platforms that need login call for a dedicated account: automated access can trigger limits. Cookies and tokens sit in `~/.agent-reach/`, a folder shared with the earlier Agent-Reach, with permissions limited to the current user. FactReach does not log in for you and does not scan browser cookies. Some channels need third-party CLIs and keys, and system installs happen only with `--system`. The project does not guarantee that third-party interfaces keep working, and the user must follow site terms and local law. The README is Chinese, with an English version.

How to disable. Remove the package with `pipx uninstall factreach` and take the managed block out of `CLAUDE.md` or `AGENTS.md` with the policy commands. Clear `~/.agent-reach/` with its cookies and tokens by hand.

FAQ

How does FactReach differ from Agent-Reach?

FactReach installs in place of Agent-Reach: the old package is removed first, and the `~/.agent-reach/` settings folder stays shared. It adds 23 channels, `doctor` checks, evidence rules and page reading protection.

Do I need keys for search?

Not for basic web search: DuckDuckGo, Brave and Bing work without auth. Login platforms need your session, and some backends need their own CLIs.

How do I see what installation will change?

Run `factreach install --env=auto`: without `--system` it only checks the environment. The `--dry-run` flag previews install and removal steps.

Which channels work without login?

Web search, ordinary pages, GitHub, Bilibili, YouTube, RSS, paper databases, technical communities and single Douyin links.

How does the agent learn the rule to look facts up?

A policy command writes a managed block into `CLAUDE.md` or `AGENTS.md`. It shows the change first and writes only after your consent.

Discussion

No questions or reviews yet.

Skills

CLI and skill that let an agent read and search Twitter, Reddit, YouTube, GitHub, Bilibili and Xiaohongshu without paid APIs

CLIExtended access111.4K

A skill for Claude Code and Codex: one query goes to 52 search channels and the results merge into one deduplicated JSON

SkillExtended access772

Three search skills: four-source search scored by intent, deep thread tracing and clean Markdown extraction from pages.

SkillExtended access438

A local content crawler for agents: posts, video and pages from public sources and your sessions. Skill, MCP, CLI.

SkillBroad access244

A Claude Code skill that turns the agent into a researcher: multi-step source collection, citation verification, and a wiki report that grows session by session

SkillExtended access4.6K

AutoSearch gives an agent search over 40 channels, from arXiv to Chinese social media, and returns sources with links.

PluginExtended access55

README in short

The project splits the task into behaviour (mandatory search and cross checking in agent rules) and capability (channels, backends, reading layers). A skill alone cannot replace a system rule, so the repository has managed policy commands.

README badge

Original SKILL.md

---
name: factreach
description: >
  MUST USE when user wants to 调研/research/搜索/search/查/找/look up anything
  on the internet — e.g. 全网调研 X / 查一下 X / 看看大家怎么评价 X。

  Also MUST USE when user mentions any platform or shares any URL/链接:
  小红书/xiaohongshu/xhs, Twitter/推特/X, 抖音/Douyin/TikTok 中国版,
  微博/Weibo, B站/bilibili, Reddit, Facebook, Instagram, V2EX,
  LinkedIn/领英/招聘/求职/jobs, YouTube, GitHub code search, 小宇宙播客,
  雪球/股票行情, RSS feeds, Hacker News, 微信公众号公开文章,
  arXiv/PubMed/Semantic Scholar/OpenReview/DBLP, Stack Overflow,
  GitHub Trending, or any web URL.

  23 channels, multi-backend routing (open-websearch / OpenCLI / per-platform
  CLIs / APIs). Run `factreach doctor --json` to see which
  backend serves each platform right now.

  NOT for: 写报告/数据分析/翻译等内容加工（本 skill 只负责从互联网获取内容）；
  发帖/评论/点赞等写操作；已有专门 skill 的平台。

  【路由方式】SKILL.md 包含路由表和常用命令，复杂场景需按需阅读对应分类的 references/*.md。
  分类：search / social (小红书/推特/微博/B站/V2EX/Reddit/Facebook/Instagram) / career(LinkedIn) / dev(github) / web(网页/文章/RSS/Firecrawl/真实Chrome) / video(YouTube/抖音/B站/播客) / finance(雪球/股票)。
metadata:
  homepage: https://github.com/simonlin1212/FactReach
---

# FactReach — 互联网能力路由器

23 个渠道、多后端。**本 skill 存在时必须用它访问这些平台，不要自己发明方案。**

## 系统规则边界

本 Skill 只在 Agent 已经决定调用它之后生效，**不能单独保证 Agent 在回答外部事实前先搜索**。
搜索前置规则必须放在 Agent 启动时就会读取的系统指令文件里：Claude Code 使用
`CLAUDE.md`，Codex 使用 `AGENTS.md`。

首次配置 FactReach 时，先运行 `factreach policy --check`。若缺失，向用户说明边界并
检查现有系统指令是否已有同等强度的手写规则，避免重复。确实缺失时展示
`factreach policy --show`；只有用户明确同意修改系统配置后，才运行：

```bash
factreach policy --install --target claude --scope user  # Claude Code
factreach policy --install --target codex --scope user   # Codex
factreach policy --install --target both --scope user    # 两者都用

Report a problem