agentleFS
TomSzenessyTomSzenessy/AI_TEMPLATE/.agents/skills/platform-reach/SKILL.md

platform-reach

Skill0 starsChanged 11 days agoMITView on GitHub

Read-only research routes for public web and the platforms agents usually improvise or give up on: web search, any page, RSS, GitHub, YouTube, Bilibili, X/Twitter, Reddit, Facebook, Instagram, LinkedIn, V2EX, XiaoHongShu and Xueqiu. Use when the user asks to research, search, look something up, or see what people say; when they share a URL; or when they name one of these platforms.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read-only: Search, list, read, and download public content.

    line 25
  2. Very specific:

    No login automation and no browser cookie-store reads: Do not drive a login form, solve or bypass a CAPTCHA, decrypt a browser profile, or …

    line 27
  3. Distinctive:

    Credentials only as environment variables, for one command: Never write a token to a file, log, issue, commit, or report.

    line 33
  4. Very specific:

    No unpinned installs: Never run npx ...@latest, uvx ...@latest, pipx install <moving-ref>, or a remote install script.

    line 36
  5. Very specific:

    Check the tool exists (command -v gh, command -v yt-dlp, ...).

    line 52
  6. Distinctive:

    Run only the read-only commands in the matching reference.

    line 54
  7. Very specific:

    Keep temporary output in /tmp/.

    line 60
  8. Very specific:

    Never use yt-dlp for Bilibili: it is blocked with 412.

    line 100

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

platform-reach

One reviewed read-only route per platform, with a fallback chain for each. The routes and commands are distilled from Panniantong/Agent-Reach at the pin recorded in resources.toml; the installer, doctor, cookie importer, update nag, Boss Zhipin CDP runbook, and Xiaoyuzhou transcription are not adopted and must not be run from here.

Boundaries

These are rules, not suggestions.

  • Read-only. Search, list, read, and download public content. Never post, reply, like, follow, message, or change an account.
  • No login automation and no browser cookie-store reads. Do not drive a login form, solve or bypass a CAPTCHA, decrypt a browser profile, or read Chrome/Safari/Firefox cookie databases. OpenCLI may reuse an existing Chrome session only when the user explicitly controls it; it does not read or export a cookie store. A manual Cookie-Editor export the user performs is their action, not this skill's.
  • Credentials only as environment variables, for one command. Never write a token to a file, log, issue, commit, or report. Export it in the child environment and do not echo it.
  • No unpinned installs. Never run npx ...@latest, uvx ...@latest, pipx install <moving-ref>, or a remote install script. If a tool is missing, name the pinned version and ask the owner before installing.
  • LinkedIn may enter only as a disabled, pinned MCP route vetted by the skill-scout role; never @latest.
  • Transcription goes to one named provider. Do not fall back to a second provider without explicit consent, because the audio leaves the machine again.
  • People are not the dataset. Collect the minimum about private individuals, never compile profiles, and follow ../../../docs/privacy.md.

Before you start

  1. Say which platform and route you will use, for example "using platform-reach: V2EX public API".
  2. Check the tool exists (command -v gh, command -v yt-dlp, ...). A missing tool is a setup question, not a reason to invent a scraper.
  3. Run only the read-only commands in the matching reference. If a route needs a login the user already has, ask them to establish it; never do it for them.
  4. On failure, follow that reference's retry chain. Stop when you have non-empty content; an exit code or a version probe is not evidence of content.
  5. Keep temporary output in /tmp/; never write research artifacts into the repository unless the brief asks for them.

Routing table

The user wantsRead
Web or code searchreferences/web.md
Any page, article, or RSS feedreferences/web.md
GitHub repositories, issues, PRs, CIreferences/dev.md
YouTube or Bilibili metadata, subtitles, commentsreferences/video.md
XiaoHongShu, X/Twitter, Bilibili, V2EX, Reddit, Facebook, Instagramreferences/social.md
LinkedIn profiles, jobs, companies, or Xueqiu quotesreferences/career-finance.md

For broad research, combine routes in parallel (a web search plus one discussion platform plus one regional platform), then synthesize. State which platforms actually returned content and which did not.

Zero-config commands

# Web search (Exa MCP, when the route is configured)
mcporter call exa.web_search_exa query="query" numResults=5

# Read any public page
curl -s "https://r.jina.ai/URL"

# GitHub search (read-only)
gh search repos "query" --sort stars --limit 10

# YouTube subtitles, no video download
yt-dlp --write-sub --write-auto-sub --sub-lang "zh-Hans,zh,en" --skip-download -o "/tmp/%(id)s" "URL"

# V2EX hot topics (public API, no auth)
curl -s "https://www.v2ex.com/api/topics/hot.json" -H "User-Agent: platform-reach/1.0"

# Bilibili search (bili-cli, read-only, no login)
bili search "query" --type video -n 5

Never use yt-dlp for Bilibili: it is blocked with 412. Use bili or OpenCLI as the reference says.

Environment and setup

There is no bundled installer or doctor. Check each tool directly:

for tool in gh yt-dlp bili opencli twitter mcporter curl; do
  command -v "$tool" >/dev/null 2>&1 && echo "$tool: present" || echo "$tool: missing"
done

A route is usable only when its tool is present and the read command returns non-empty content. When a tool is missing or a login is absent, say which route is unavailable and offer the fallback in that reference; do not install anything without the owner's approval.

The pinned upstream record and its linked Boss Zhipin/Xiaoyuzhou files are listed by make upstream ID=agent-reach; pull one of those files only when the need named in its when trigger is real.

Special cases

  • OpenCLI uses a Chrome session the user already controls. If none exists, do not automate a login; report the route unavailable.
  • XiaoHongShu needs the full URL or ID from a search result (the xsec_token is part of it); a bare note id is not a valid route.
  • Reddit has no anonymous path. Both routes need an existing login; do not recommend creating API credentials for a new user.
  • X/Twitter takes TWITTER_AUTH_TOKEN and TWITTER_CT0 in the child environment only. A search may fail even when the timeline works; follow the retry chain.
  • Boss Zhipin and Xiaoyuzhou are deliberately out of scope. Their upstream runbooks are linked from resources.toml, not copied here, because they require login automation or a transcription provider decision.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.