SuperAPI AI Directory

agent-reach

General & other

View repository

Updated 11-10 16:19

About this skill

Agent Reach

Give your AI agent eyes to see the entire internet. Read & search Twitter/X, Reddit, YouTube, GitHub, LinkedIn, Bilibili, and login-walled web platforms with one CLI and zero API fees.

Overview

One command for every platform: agent-reach. Read-only; returns structured JSON output.

agent-reach channels                  # List available channels, actions, and parameters
agent-reach <channel> <action> [args] # e.g. agent-reach github search "vulnerability scanner"
agent-reach read <url>                # Automatically select channel and read any URL

🔌 Supported Channels

Tier Channel Capabilities Source / Upstream Tools
Default web General webpage reading, content extraction, search fallback Jina Reader, Exa
Default youtube Video transcripts, subtitles, audio speech-to-text yt-dlp, whisper
Default github Repositories, files, issues, commit logs, pull requests GitHub public APIs
Default bilibili Video details, subtitles, Danmaku comments bili-cli, public endpoints
Default v2ex Topic reading, community discussions V2EX public feeds
Optional reddit Subreddit search, threads, comment trees Reddit json/RSS
Optional linkedin Posts, company profiles, job postings OpenCLI / CloakBrowser
Optional twitter / x Tweets, search, user profiles OpenCLI / CloakBrowser
Optional xiaohongshu Note descriptions, tags, comments CloakBrowser session
Optional boss Developer job listings, skill demands boss-agent-cli
Optional xueqiu Stock quotes, financial sentiment Xueqiu public endpoints
Optional xiaoyuzhou Podcast transcripts, summaries Groq Whisper API

⚡ Integration with REA (Reverse Engineer Anything)

When combined with the REA suite:

  1. Threat & Vulnerability Research: When auditing a website or binary with REA, use Agent Reach to instantly look up CVEs, GitHub discussions, and security advisories without leaving your terminal:
    agent-reach github search "package-name vulnerability CVE"
    agent-reach reddit search "framework-name security breach"
    
  2. Deep Web Reading: When a website or API blocks standard curl/fetch with Cloudflare or bot protection, use:
    agent-reach read https://target-app.com/docs
    
  3. Tutorial & Exploit Video Transcripts: Extract technical explanations directly from YouTube or Bilibili video transcripts without watching the video:
    agent-reach youtube read "https://www.youtube.com/watch?v=..."
    

🚀 Setup & Installation

Install Agent Reach with uv (recommended):

uv tool install https://github.com/Panniantong/Agent-Reach/archive/refs/tags/v1.6.0.zip
agent-reach setup

To manage channels and dependencies:

agent-reach add <channel>      # Install dependencies for an optional channel
agent-reach remove <channel>   # Remove an unused channel
agent-reach doctor --text      # Check system dependencies and browser status
agent-reach check <channel>    # Verify connectivity for a specific channel
agent-reach update             # Upgrade Agent Reach and installed channel tools

📊 Result Structure & Error Handling

Outputs are standard JSON with { ok, channel, action, backend, items, meta, error }:

  • ok: true: Results found in items (title, url, author, published_at, text, raw).
  • ok: false: Examine error.code:
error.code Meaning Action to Take
NEED_SETUP Dependency missing Run agent-reach add <channel>. If key is needed, user runs agent-reach configure <key>
NEED_LOGIN Site requires login User runs agent-reach login <channel> in their own window. (Never log in for them)
UPSTREAM_BROKEN Platform changed layout Inform user; run agent-reach update if agreed
TIMEOUT Request timed out Run agent-reach doctor --text; do not retry in an aggressive loop
RATE_LIMITED Platform rate limit Wait retry_after_seconds specified in meta
PLATFORM_BLOCKED Captcha / Cloudflare wall User opens official page to solve captcha
NOT_FOUND Content nonexistent Check target URL or keyword

🛡️ Security & Red Lines

  • Read-Only: Agent Reach never posts, comments, likes, or sends messages.
  • Credential Safety: Never logs in for the user; never asks for, captures, or logs passwords.
  • Dedicated Profile: Browser sessions run inside an isolated profile under ~/.agent-reach/browser/profile and shut down after 10 idle minutes.