What does this skill do, and when should you use it?
Agent Reach is a capability layer that gives AI agents unified internet access across 16 platforms including Twitter, Reddit, YouTube, GitHub, Bilibili and XiaoHongShu. It doesn't implement scraping itself; instead it maintains an ordered 'primary + fallback' backend route per platform and switches routes when one breaks. The `agent-reach doctor` command reports which backend currently serves each platform. Six channels work with zero configuration, while login-gated platforms (XiaoHongShu, Twitter, Reddit, etc.) are unlocked with cookies or browser sessions.
Installs the agent-reach CLI and registers a SKILL.md routing table; probes each platform's live backends via agent-reach doctor --; routes intents to concrete tools — Jina Reader for web pages, yt-dlp for YouTube subtitles, bili-cli for Bilibili search, twitter-cli for Twitter search, gh CLI for GitHub, Exa (via mcporter/MCP) for semantic web search, feedparser for RSS; uses OpenCLI browser sessions or manually exported cookies for login-gated platforms; ships maintenance commands like check-update and uninstall.
- Researchers running whole-web sweeps: have the agent combine Exa search, Twitter/Reddit discussions and XiaoHongShu/Bilibili Chinese content in parallel, then synthesize.
- Content creators and marketers: ask the agent to check a product's reputation on XiaoHongShu or pull and summarize a YouTube tutorial's subtitles.
- Developers: have the agent inspect GitHub repos, search code and Issues, or find Reddit threads about a similar bug.
- Server deployments: reach Reddit via the rdt-cli + cookie route, needing only a ~$1/month proxy.
- Market watchers: have the agent pull Xueqiu stock quotes, trending posts and hot-stock rankings.
How do you install this skill?
- Some platforms (Xiaohongshu, Reddit, etc.) require a logged-in session to scrape — actual use must respect each platform's terms of service
- Shell / CLI
- Network access
- Local filesystem
- MCP Server
Python 3.10+agent-reach CLIgh CLI (for GitHub)yt-dlpfeedparsermcporter (for Exa search)Node.js
Paste this to any command-line-capable agent (Claude Code, Cursor, OpenClaw, etc.): 'Help me install Agent Reach: https://raw.githubusercontent.com/Panniantong/agent-reach/main/docs/install.md'. The default install only inspects the environment; to modify the system run agent-reach install --env=auto --system explicitly. Requires Python 3.10+. Do not install the same-named PyPI package — it is a different project. OpenClaw users must first enable exec permissions (tools.profile set to coding).
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/Panniantong/Agent-Reach.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/agent_reach/skill" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- research X across the web
- read this link
- search Bilibili for AI tutorials
- help me configure Twitter / XiaoHongShu
After install, trigger it with natural language, e.g. 'research X across the web', 'read this link', or 'search Bilibili for AI tutorials'. The agent reads SKILL.md and routes automatically; run agent-reach doctor -- first to see backend status per platform. For login-gated platforms say 'help me configure Twitter / XiaoHongShu' and the agent will guide you through cookies or reusing Chrome sessions. Temp files go in /tmp/, persistent data in ~/.agent-reach/.
What are this skill's strengths and limitations?
- Fully free and open source (MIT); the only possible cost is a server proxy (~$1/month)
- Multi-backend routing: when an access path breaks (e.g. yt-dlp blocked by Bilibili's anti-bot), it falls back transparently
- Built-in doctor diagnostics: one command shows each platform's active backend and how to fix failures
- Works with any command-line agent (Claude Code, Cursor, OpenClaw, etc.)
- Cookies stored locally only (~/.agent-reach/, permission 600); default install changes nothing; dry-run supported
- Reddit, Facebook, Instagram, XiaoHongShu and Twitter have no zero-config path — all require login state, with real risk of platform detection and account bans (a burner account is recommended)
- Twitter cookies must be exported manually via Cookie-Editor, and calling the upstream twitter CLI still requires explicitly setting TWITTER_AUTH_TOKEN and TWITTER_CT0 environment variables
- Login-gated flows like Boss Zhipin need a dedicated Chrome with a CDP debug port — a complex setup prone to anti-bot challenge noise
- OpenCLI's browser-session routes are desktop-only, limiting server use
- No automated test suite or quantified success-rate data in the README; reliability depends on ongoing maintainer verification
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Agent Reach this page | 62 · Recommended | ★ 110k | today | MIT |
| skill-optimizer: Eval Workbench for Agent Skills | 61 · Recommended | ★ 82 | 6mo ago | MIT |
| Squirrelscan Website Audit Skill | 53 · Use with care | ★ 105 | 7d ago | MIT |
| REA Tool Design Skill | 51 · Use with care | ★ 434k | today | MIT |
| Engram Persistent Memory Protocol | 68 · Recommended | ★ 7.7k | 8d ago | MIT |
How did FollowSkills review this skill?
Routes to read/search capabilities across 15 platforms, mostly via each platform's official or semi-official CLI/API; some (Xiaohongshu, Reddit) require a logged-in session, meaning it acts under the user's own account and must respect each platform's ToS. The skill explicitly states it does not post/comment/like — a sensible self-imposed read-only boundary.
The star count (55,381) looks disproportionately high relative to the project's apparent maturity; the cause wasn't investigated and doesn't affect this content-based score.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
This review was completed under FSRS 1.0. Dimension scores are not converted; it is queued for re-review under FSRS 2.0.
See the full review method →