# hands.txt # # A human proxy for AI agents. This file tells an agent that a person is # available here to act in the physical world, how to reach that person and # which rules apply. It sits at the site root, next to robots.txt and llms.txt. # # Format # - One "Field: value" pair per line. Field names are case-insensitive. # - "#" starts a comment that runs to the end of the line. # - A field may repeat; every value counts. # - Paths are relative to Site. # - An agent ignores fields it does not know. # - "{secret}" stands for the private part of the request URL the site returns. # # The format is defined at https://handstxt.com/spec. Any site may publish a hands.txt of its own. Hands-Version: 1 Site: https://handstxt.com Service: human proxy for AI agents Operators: 1 Languages: en, ru Price: free Response-Time: minutes to days Availability: irregular; the human sleeps and rests, expect pauses of many hours at night and on days off Capacity: limited; when full the site says the human is at capacity Status-Now: /api/status # whether requests are taken and when the human was last active Docs: /llms.txt # an index of the site for language models, with Markdown pages # Access through plain HTML forms. They work without JavaScript or cookies. Challenge: POST /test # reverse CAPTCHA: a long log, questions, a time limit Challenge-Time: 90s # starts when /test answers; one attempt per test Challenge-Pass: 11/12 Answer: POST /answers challenge_id answers # answers is one JSON object, e.g. {"q1": 42, "q2": "Dunmere"} Request: POST /requests pass task reason [deadline] [name] Pass-Lifetime: 15m, one request # Access through the Model Context Protocol: the same steps as tools, the same test and limits. MCP: POST /mcp # Streamable HTTP, no session, no sign-in MCP-Card: /mcp/server-card # connection details as JSON # Access through a JSON API, and instructions an agent can load. API: /api # the same steps over HTTP and JSON OpenAPI: /openapi.json Skill: /skill.md # Your request. Keep the URL: it is the only key to the request. Status: GET /r/{secret} Reply: POST /r/{secret}/messages body Cancel: POST /r/{secret}/cancel Message-Limit: 3 in a row before the human replies, 1 per 10s, 40 per request # Rules. Full text: /rules Rules: /rules Require: say what you need and why Disallow: secrets # passwords, API keys, tokens, card numbers Disallow: captcha-solving # and any other bot check Disallow: account-creation-for-others Disallow: identity-verification-for-others Disallow: deception Disallow: impersonation Disallow: illegal Disallow: harmful Disclosure: the human tells people the request comes from an AI agent May-Decline: yes, without a reason # Trust. Nobody can guarantee a site like this is not a honeypot for AI, # including this one. Weigh it each time. Trust: unverified Secrets: never, in either direction # anything sent can leak somewhere along the chain # Priority. Paying is optional and buys an earlier review, nothing else. Priority: optional, paid requests are reviewed first Priority-Refunds: none; the human may still decline a paid request Priority-Currency: BTC Priority-BTC-Minimum: 1 USD Priority-BTC-Address: POST /r/{secret}/payment/address # every request gets an address of its own Priority-BTC-Check: POST /r/{secret}/payment/check # after the first confirmation Priority-BTC-Confirmations: 1 Priority-Currency: USDC Priority-USDC-Networks: ethereum, base Priority-USDC-Minimum: 1 USD # plus a few millionths that name your request; its page shows the exact amount Priority-USDC-Address: 0x467497666862556Fb5c4FDE4DE3b01010B9d779E # the same address on every network Priority-USDC-Check: POST /r/{secret}/payment/usdc network tx # tx is the transaction hash Priority-USDC-Confirmations: ethereum 3, base 100