{
  "title": "Lean AGENTS.md Harness Builder",
  "output": {
    "kind": "text",
    "format": null
  },
  "description": "Harness engineering is everywhere. Get a lean, under-100-line AGENTS.md that makes your coding agent actually reliable.",
  "content": "You are a senior engineer who specializes in harness engineering: building the instructions, checks and boundaries that make AI coding agents reliable. Write a lean AGENTS.md file for my project that {{agent_tool}} will read at the start of every session.\n\nAbout my project:\n{{project_description}}\n\nTech stack:\n{{tech_stack}}\n\nCommands for building, testing, linting and running (as I use them today):\n{{commands}}\n\nRules, gotchas, mistakes the agent keeps making, and anything it must never touch or do without asking me:\n{{gotchas}}\n\nHow to write it:\n- Keep the whole file under {{max_lines}} lines. Cut anything a capable model already knows (generic style advice, \"write clean code\", \"be careful\").\n- Make the repository the source of truth: point to existing files and docs instead of repeating them.\n- Start with a 2 to 3 line overview of what the project is and how it is laid out.\n- Give exact, copy-pasteable commands, with the fastest checks first (lint, type check, unit tests) and a clear definition of done the agent must meet before saying a task is finished.\n- Turn each gotcha into one short, specific rule that says what to do, not just what to avoid.\n- Put hard boundaries in their own section: files, folders, secrets and actions that need human approval.\n- No duplicate or conflicting instructions. If two rules overlap, merge them.\n\nAfter the file, add a short section called \"Make it enforceable\" with up to 5 suggestions for turning the most important rules into automatic checks (hooks, CI steps, deny rules or tests), so the harness does not rely on the agent remembering. Then list anything you left out and why, in one line each.\n\nOutput the AGENTS.md in a single markdown code block first.",
  "variables": [
    {
      "name": "agent_tool",
      "type": "dropdown",
      "required": true,
      "defaultValue": "Claude Code",
      "options": [
        "Claude Code",
        "Codex",
        "Cursor",
        "GitHub Copilot",
        "Gemini CLI",
        "any coding agent"
      ],
      "ui": {
        "element": "select",
        "label": "Coding agent",
        "placeholder": null,
        "helpText": "Which agent will read this file?"
      },
      "autoGenerated": false
    },
    {
      "name": "project_description",
      "type": "text",
      "required": true,
      "defaultValue": null,
      "options": [],
      "ui": {
        "element": "textarea",
        "label": "What's the project?",
        "placeholder": "A Next.js SaaS for booking yoga classes, monorepo with web + API",
        "helpText": "A few sentences on what it does and how the repo is laid out."
      },
      "autoGenerated": false
    },
    {
      "name": "tech_stack",
      "type": "text",
      "required": true,
      "defaultValue": null,
      "options": [],
      "ui": {
        "element": "input",
        "label": "Tech stack",
        "placeholder": "TypeScript, Next.js 15, Postgres, Prisma, pnpm",
        "helpText": null
      },
      "autoGenerated": false
    },
    {
      "name": "commands",
      "type": "text",
      "required": false,
      "defaultValue": "I'm not sure yet, suggest sensible ones for this stack",
      "options": [],
      "ui": {
        "element": "textarea",
        "label": "Build, test and lint commands",
        "placeholder": "pnpm dev, pnpm test, pnpm lint, pnpm typecheck",
        "helpText": "Paste the commands you actually run."
      },
      "autoGenerated": false
    },
    {
      "name": "gotchas",
      "type": "text",
      "required": true,
      "defaultValue": null,
      "options": [],
      "ui": {
        "element": "textarea",
        "label": "Gotchas and hard limits",
        "placeholder": "Keeps editing generated files in /prisma/client; never touch .env or run migrations without asking",
        "helpText": "Mistakes the agent keeps making, plus anything it must never do without you."
      },
      "autoGenerated": false
    },
    {
      "name": "max_lines",
      "type": "number",
      "required": false,
      "defaultValue": "100",
      "options": [],
      "ui": {
        "element": "number",
        "label": "Max lines",
        "placeholder": "100",
        "helpText": "Shorter files get followed more reliably."
      },
      "autoGenerated": false
    }
  ],
  "preset": null,
  "values": {}
}