What is SkillRouter

What SkillRouter is

Models can write code and read docs, but they stall on real websites: selectors have to be guessed, logins have to be wired up, and every redesign breaks what worked yesterday. SkillRouter turns that trial-and-error into deterministic commands once, and takes responsibility for keeping them working.

Each skill = one Product Contract (YAML — the single source of truth for price, input, output and refund rules) + one client-side execution adapter. Accounts, checkout, entitlements, claims and daily verification are shared platform services; business execution stays on the buyer's device.

What you are buying

Not data, and not a black box that acts on your behalf.

• Deterministic commands — one line like sr xiaohongshu search "agent tools", with zero LLM calls and zero tokens at run time.
• A manual your agent reads itself — you learn nothing new; sr manual <skill> points the agent at it.
• Proof it still works today — every skill carries a verified_at stamp from a real-machine check.
• A platform-level promise — broken skills are pulled from sale and confirmed unrepairable products are refunded, instead of a per-product disclaimer.

Two kinds of skill

Two things sit on the shelf; they are called and billed differently.

What you can do with it

• Content operations: read and publish chains for XiaoHongShu / Bilibili / Douyin, running inside your own creator-console session.
• Job hunting: search roles, read JDs and filter opportunities across LinkedIn / BOSS Zhipin / 51job / Indeed.
• Public read-only data: EastMoney quotes, Jianyu tender notices, policy and statute search, arXiv papers.
• Back-office ops: read-only Stripe reconciliation (bring your own restricted key, zero write endpoints).
• Chinese model web consoles: DeepSeek / Doubao / Kimi / Qwen / Wenxin / Yuanbao / Zhipu — through the web account you already have, no separate API purchase.

Who it is for

• People working in Claude Code / Codex / Cursor who keep hitting “my agent can't open that site”.
• People who need an agent to repeat actions inside a logged-in session, without handing the account to a third-party host.
• People willing to pay so they never have to watch a platform's redesigns themselves.

Set up in three minutes

One line to install; hand the rest to your agent.

npm i -g @skill-router/cli
sr connect skillrouter.org --kit default   # approve in the browser, key saved for you
sr setup                                   # runtime + browser Bridge + purchased skills
sr integrate                               # wire into Claude Code / .agents

• sr catalog lists the shelf; sr manual <skill> points your agent at the manual.
• sr setup --check diagnoses only — no installs, no waiting, no login.
• Agents can also connect over MCP: npx @skill-router/mcp.
• Machine-readable entry points: /SKILL.md and /llms.txt.

Freshness and refunds

Free open-source adapters solve “does it exist”. SkillRouter solves “does it still work today, and who pays when it doesn't”. That is the only thing we charge for.

• Daily real-machine checks: assertions run against the live site on a schedule; verified_at only advances on a pass. That timestamp is what you see on the site.
• Repeated failures pull the skill from sale automatically — we do not keep selling a broken path.
• Within 7 days of delivery, you can file a claim for a stale or materially incorrect product; confirmed issues we cannot repair receive a full refund.
• If a skill breaks during your membership it comes off the shelf — you never pay for a broken one.

Boundaries: what we do not do

• We do not host your accounts, log in on your behalf, or clear QR/CAPTCHA challenges for you.
• Execution stays on your client; our servers never visit third-party sites for you.
• We do not bypass any site's anti-bot or human-verification measures; a challenge is reported as an error, not worked around.
• We do not repackage a platform's data endpoints as a product.
• Your use of third-party platforms remains governed by their terms and the account risk is yours; use a dedicated account for session-based commands.

Compared with building it yourself

Next step