Agent Reach is a capability layer, not another scraper. It does not read anything itself. For each platform it keeps an ordered list of backends (a first choice plus fallbacks), installs the working one, and runs a health check so your agent knows what it can currently reach. Your agent then calls those upstream tools directly.
That distinction matters for a boring but important reason: access paths break constantly. The repo's own example is Bilibili 412-blocking yt-dlp in June 2026, which moved that platform onto a different backend with no action from users. You are installing the router, not betting on one scraper staying alive.
agent-reach doctor, prints which platforms are live, which need a step from you, and what that step is.This is the part the viral reels skip. "Install it once and just ask" is true for some platforms and not others, and knowing which is which saves you an hour of confusion.
You do not run an installer yourself. You hand your agent the install document and it does the work, which is the actual novelty here.
python3 --version in a terminal tells you) and an agent that is allowed to run shell commands. Claude Code, Cursor and Windsurf all qualify out of the box.--system, read what it wants to install before agreeing.agent-reach doctor. You get a list like "6/9 channels available" with a tick next to what works and a fix prescription next to what does not. This is your source of truth, not a blog post.npx skills add Panniantong/Agent-Reach@agent-reach. The skill teaches your agent which command to reach for, so you never memorise CLI syntax.Install Agent Reach: https://raw.githubusercontent.com/Panniantong/agent-reach/main/docs/install.md
Read that document fully before you do anything, then follow it. Do the read-only environment check first and show me exactly which packages and system changes you would need before you make any of them, because I want to approve system-level installs myself rather than have them happen silently. When you are finished, run agent-reach doctor and show me the raw output so I can see which platforms are live and which still need a step from me.Installing it is not the point. This is the point. Each of these is a real research job that Claude could not do before, and each one is written to make it show you evidence rather than a confident summary.
Use Agent Reach to search Reddit for what people actually complain about regarding [your topic, product category or niche], across the subreddits where that audience really posts rather than the obvious big one.
Pull at least 30 posts or comments from the last six months, and quote the complaints verbatim in their own words rather than paraphrasing them into marketing language, because the exact phrasing is the thing I want. Group them into recurring themes, rank the themes by how often they come up and how emotional the language is, and note which ones already have a good answer in the thread versus which ones nobody solves. For every theme, give me the single best verbatim quote and a link to the thread it came from so I can verify it myself.Use Agent Reach to pull the full transcript of this YouTube video: [URL]. Then pull the transcript of the two or three other videos on that channel with the highest view counts.
Break each one down structurally rather than summarising it: what the first fifteen seconds do to earn attention, how the body is sequenced, where the pace changes, what specific claims or numbers are used as proof, and how it closes. Then tell me what is repeatable across all of them versus what only worked because of that specific topic. Finish with the three structural moves I could apply to my own content this week, and be blunt about which parts of their approach would not suit me.Use Agent Reach to search Twitter/X for reactions to [topic, product launch or announcement] over the last two weeks, and read the replies to the highest-engagement posts, not only the posts themselves.
Map the argument: what are the two or three genuinely distinct positions people hold, who is making each one, and what evidence does each side actually cite. Separate the takes that are informed from the ones that are just repeating a headline, and flag any factual claim that keeps getting repeated without a source. Then tell me which angle is already saturated and which one is underserved, and quote the specific posts you drew that from with links.Use Agent Reach to set me up a weekly research loop on [your niche]. Find the five or six most useful RSS or Atom feeds and the two or three most active GitHub repositories in this space, and confirm each one actually returns content before you include it.
Once you have them, read the last week of each and write me one brief: what genuinely changed, what is just noise, and what it means specifically for someone doing [your work]. Keep the same structure every week so I can compare briefs at a glance, and where you are inferring significance rather than reporting something stated outright, label it so I can tell the difference. Save the feed list somewhere reusable so the next run does not start from scratch.agent-reach doctor first, then update with the same one-line pattern (swap install.md for update.md in the URL). Most breakages are a re-route, not a dead end.
Agent Reach on GitHub (MIT, the source of every claim above)
The official install document your agent reads
yt-dlp, Jina Reader, gh CLI and OpenCLI, the backends doing the actual reading
Claude Code skills documentation, if you install it the skill way