mechanical-turk: Instructions file for Claude Code

CLAUDE.md

mechanical-turk CLAUDE.md is an instructions file for Claude Code from hobsojam/mechanical-turk. It costs 618 tokens per session, scanned A, original, MIT.

Project instructions for Mechanical Turk, a human-operator console that AI agents connect to when they need a person to make a decision. The project uses MCP, a standard way for AI tools to communicate with software.

In plain words
What is it for?
Understanding or changing the server, browser console, MCP tools, question queue, build commands, and manual testing process.
Why use it?
They give the coding agent the project layout, data flow, development commands, and current testing situation, so it can work in the expected way.

Instructions file for Claude Code

Written for Claude Code: the file is CLAUDE.md.

This is hobsojam/mechanical-turk's own configuration. It tells Claude Code how to work on mechanical-turk itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything mechanical-turk configures →

Reuse

Borrowing it

Nothing to install: this file belongs to hobsojam/mechanical-turk. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/hobsojam/mechanical-turk/main/CLAUDE.md
Clone the repo
git clone --depth 1 https://github.com/hobsojam/mechanical-turk

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for mechanical-turk CLAUDE.md

README.md
[![agentmods](https://agentmods.dev/badge/instructions/hobsojam/mechanical-turk/claude-md/github.svg)](https://agentmods.dev/instructions/hobsojam/mechanical-turk/claude-md)
Your own site
<a href="https://agentmods.dev/instructions/hobsojam/mechanical-turk/claude-md"><img src="https://agentmods.dev/badge/instructions/hobsojam/mechanical-turk/claude-md/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for mechanical-turk CLAUDE.md

Your own site · 80×15
<a href="https://agentmods.dev/instructions/hobsojam/mechanical-turk/claude-md"><img src="https://agentmods.dev/badge/instructions/hobsojam/mechanical-turk/claude-md.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 618 This file is loaded in full into every session.
When invoked 618 The same file — it is already loaded in full.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00618 $0.00618
Opus 5.5 $0.00247 $0.00247
Sonnet 5.5 $0.00124 $0.00124
Haiku 4.5 $0.00062 $0.00062

Measured 2mo ago against content hash 92651eecd860, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

mechanical-turk CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2mo ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

CLAUDE.md · 54 lines

How it starts

The opening of the file, as written. The whole thing — 54 lines — stays where its author put it; the contents beside it link to each section on GitHub.

mechanical-turk

Project Overview

Human-operator console that AI agents connect to via MCP (Streamable HTTP transport). The inversion of a chatbot: the AI drives the conversation, the human is an on-call resource. When an agent needs a human decision it calls ask_human, which blocks until the operator answers in the browser console.

Architecture

src/
  index.ts   — Express server, WebSocket hub, MCP route handlers, session registry
  mcp.ts     — McpServer factory (one server per agent session); defines ask_human + notify_human tools
  queue.ts   — EventEmitter-based question queue; bridges MCP tool calls to browser answers
public/
  index.html — Single-file browser console (vanilla JS, WebSocket, no build step)

Data flow: MCP tool call → queue.add() → Promise held open → queue.on("question") event → broadcast to browser → human submits answer → queue.answer() → Promise resolves → MCP tool returns.

Development Commands

npm install    # install dependencies
npm run dev    # tsx watch — restarts on any src/ change
npm run build  # tsc → dist/
npm start      # node dist/index.js

Testing

There is no automated test suite yet. To verify manually:

  1. npm run dev — server should print the localhost URL
  2. Open http://localhost:3000 — browser console should load
  3. Configure an MCP client to point at http://localhost:3000/mcp and call ask_human — question should appear in the browser inbox

Key decisions

  • One McpServer per session: each connecting agent gets its own McpServer instance so agentId is captured in closure. Sessions are stored in a Map<sessionId, { transport, agentId }>.
  • EventEmitter on queue: avoids coupling between mcp.ts and the WebSocket layer. index.ts wires the events.
  • Vanilla HTML: no framework in public/index.html — keeps the UI dependency-free and the whole thing a single file.
  • rejectByAgent on disconnect: when an agent disconnects, its pending questions are rejected so the AI gets an error rather than hanging forever.

Read the full file on GitHub · 54 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2mo ago First seen · 54 lines · 618 tokens per session scan A 92651eecd860

Subscribe to this mod's changes

mechanical-turk CLAUDE.md is an instructions file published in the GitHub repository hobsojam/mechanical-turk (0 stars, last pushed 5d ago), licensed MIT. It adds 618 tokens to every session, about $0.0025 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

mechanical-turk CLAUDE.md compared with instructions from other repositories

mechanical-turk CLAUDE.md next to 6 instructions from other repositories
ModTypeStarsLast pushLicenceScanTokensWorks with
mechanical-turk CLAUDE.mdhobsojam/mechanical-turkInstructions file05d agoMITA618Claude Code
vscode buildNext.instructions.mdmicrosoft/vscodeInstructions file196ktodayMITA9,719GitHub Copilot
next.js AGENTS.mdvercel/next.jsInstructions file144ktodayMITA8,130Codex, OpenCode
vscode oss-third-party-notices.instructions.mdmicrosoft/vscodeInstructions file196ktodayMITA5,685GitHub Copilot
langchain AGENTS.mdlangchain-ai/langchainInstructions file150ktodayMITA4,680Codex, OpenCode
deepseek-harness AGENTS.mddeepseek-ai/deepseek-harnessInstructions file283ktodayMITA4,342Codex, OpenCode
vscode copilot-instructions.mdmicrosoft/vscodeInstructions file196ktodayMITA3,153GitHub Copilot