Support: fuel the next build:

Direct answer

What is agentprobe? An open-source, local-first CLI auditor and CI quality gate for the agentic web. Google Lighthouse tests web performance for humans; agentprobe tests whether autonomous AI agents (Claude, ChatGPT, Perplexity, OpenAI Operator, and headless browser bots) can reach, parse, and take actions on your site without getting blocked or burning excessive tokens.

Who is it for? Frontend developers, full-stack engineers, and open-source maintainers who want their documentation, apps, and platforms to be accurately cited and executable by autonomous AI agents.

What does it cost? Free. MIT licensed. Runs 100% locally on your machine with zero API keys, zero accounts, and zero cloud lock-in.

How fast? Audits live URLs or local ./dist static builds in under 2 seconds.

$ npx @akashpriyadarshii/agentprobe https://akashpriyadarshi.is-a.dev

  ┌──────────────────────────────────────────────────────────────┐
  │                    agentprobe v0.0.2                         │
  │           The Open-Source Lighthouse for AI Agents           │
  └──────────────────────────────────────────────────────────────┘

  Target:      https://akashpriyadarshi.is-a.dev/
  Audited:     2026-10-06T20:10:30.096Z
  Gate Status: PASSED (min score: 80)

  ════════════════════════════════════════════════════════════
  TOTAL READINESS SCORE: 85 / 100
  [██████████████████████████░░░░] 85%
  ════════════════════════════════════════════════════════════

Why agentprobe?

Websites designed only for human eyeballs break when autonomous agents attempt to interact with them:

• Accidental AI crawler lockouts: Over-aggressive WAF firewalls or blanket robots.txt disallows silently block GPTBot, ClaudeBot, or PerplexityBot from indexing your site.
• Token bleed from messy DOM trees: Agents scraping bloated, un-semantic HTML burn tens of thousands of tokens per page request, causing reasoning models to lose context.
• Empty client-side hydration traps: Single-page applications (SPAs) that serve empty <div id="root"></div> shells blind static AI crawlers that do not execute heavy client-side JavaScript.
• Dead-end interactive controls: Icon-only buttons without text or aria-label prevent autonomous browsing agents from clicking navigation items, executing searches, or submitting forms.

agentprobe gives you an empirical 0-100 Agent Readiness Score with a prioritized checklist of fixes before you deploy.

Quickstart

Install & Run

Run instantly via npx:

npx @akashpriyadarshii/agentprobe https://docs.yourdomain.com

Or install globally to use the agentprobe command anywhere:

npm install -g @akashpriyadarshii/agentprobe
agentprobe https://docs.yourdomain.com

1. Audit a local production build before deploying

agentprobe ./dist --min-score 85

2. Auto-scaffold missing AI manifests

agentprobe init

Generates spec-compliant llms.txt, a 15-bot permissive robots.txt, and discovery manifests (.well-known/ai-catalog.json, .well-known/mcp.json) directly in your public/ or root directory.

Shell Aliases

# bash/zsh (~/.bashrc or ~/.zshrc)
alias probe='npx @akashpriyadarshii/agentprobe'

# PowerShell ($PROFILE)
Set-Alias probe agentprobe

The 5 Audit Pillars

agentprobe evaluates websites across five weighted pillars (20 points each, 100 points total):

| Pillar | Weight | What it inspects | Why it matters for AI Agents | |---|:---:|---|---| | 1. Bot Accessibility | 20 pts | robots.txt permissions (15 AI crawlers), WAF blocks, XML sitemap | Guarantees major models (GPTBot, ClaudeBot, PerplexityBot, Amazonbot, ByteSpider) can crawl. | | 2. LLM Manifest (llms.txt) | 20 pts | /llms.txt, /llms-full.txt, H1/blockquote syntax, token budget | Gives agents a lightweight, structured markdown summary of your site, saving 90%+ token costs. | | 3. Structured Context | 20 pts | Schema.org JSON-LD, HTML5 landmarks (<main>, <nav>), headings | Eliminates hallucination by providing unambiguous, machine-readable entity metadata. | | 4. Actionability & Negotiation | 20 pts | Accept: text/markdown support, button aria-labels, SSR vs SPA shells | Ensures agents can execute form submissions and clicks without stumbling on unlabeled DOM nodes. | | 5. Protocol Discovery & Commerce | 20 pts | /.well-known/ai-catalog.json, /.well-known/mcp.json, skills index, x402 payment headers | Enables autonomous tool discovery (ARD, MCP) and agentic commerce / micropayment protocols. |

CLI Reference

agentprobe [OPTIONS] <TARGET>

| Option | Default | Description | |---|:---:|---| | <TARGET> | Required | URL (https://... or http://localhost:3000) or static directory (./dist). | | --min-score <n> | 80 | Minimum passing score threshold (0-100) for CI quality gates. | | --plain | false | Outputs structured plain text without ANSI codes for AI agents and scripts. | | --json | false | Outputs machine-readable JSON to stdout for programmatic analysis. | | --summary | false | Outputs GitHub Step Summary Markdown for pull request checks. | | --no-color | false | Disables ANSI colors in terminal output (respects NO_COLOR=1). | | -q, --quiet | false | Suppresses progress banners; outputs final exit code and report only. | | -v, --version | None | Prints current agentprobe version. | | -h, --help | None | Displays CLI help synopsis. |

Commands

| Command | Description | |---|---| | agentprobe <target> | Audit a live URL or static build folder. | | agentprobe init | Interactively scaffold a compliant llms.txt and permissive robots.txt. |

Exit Codes

• 0: Audit passed (score ≥ threshold).
• 1: Audit failed (score < threshold).
• 2: Invalid target, unreachable network, or parse failure.

CI/CD & GitHub Action

Block pull requests that introduce regressions in agent readability or block AI crawlers:

name: Agent Readiness Gate

on: [push, pull_request]

jobs:
  agent-audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Audit Preview Deployment
        uses: AkashPriyadarshii/agentprobe@v1
        with:
          target: 'https://preview.yourdomain.com'
          min-score: '85'

Or via direct script:

npx agentprobe ./dist --min-score 80 --json > agent-report.json

Dynamic Badge

Every audit generates a markdown badge you can paste into your project README:

[![Agent Readiness](https://img.shields.io/badge/Agent_Readiness-98%2F100-brightgreen?style=flat-square)](https://github.com/AkashPriyadarshii/agentprobe)

Architecture

agentprobe/
├── src/
│   ├── cli.ts                 # CLI entry point (commander)
│   ├── index.ts               # Core module exports
│   ├── types/index.ts         # Diagnostic & scoring contracts
│   ├── core/
│   │   ├── runner.ts          # Parallel multi-target engine executor
│   │   └── scorer.ts          # Deterministic deduction-based scoring
│   ├── engines/
│   │   ├── crawler.ts         # robots.txt & 15 AI bot permissions
│   │   ├── llmstxt.ts         # llms.txt & llms-full.txt parser + token budget
│   │   ├── semantic.ts        # Schema.org JSON-LD & HTML5 landmark validator
│   │   ├── actionability.ts   # Markdown content negotiation & ARIA accessibility
│   │   └── protocols.ts       # Protocol discovery (ARD, MCP & x402 commerce)
│   ├── reporters/
│   │   ├── terminal.ts        # ANSI high-contrast terminal UI with grade badges
│   │   ├── plain.ts           # Structured plain text for AI agents and scripts
│   │   ├── json.ts            # Machine-readable JSON
│   │   ├── summary.ts         # GitHub Actions $GITHUB_STEP_SUMMARY
│   │   └── badge.ts           # Dynamic Shields.io badge generator
│   └── scaffolder/
│       └── init.ts            # llms.txt, robots.txt, and .well-known scaffolder
├── tests/                     # 17 passing Vitest unit tests
├── action.yml                 # Composite GitHub Action manifest
└── README.md

Ecosystem

Built by Akash Priyadarshi as part of the autonomous developer tools ecosystem:

• akash-design-engineering • Apex meta-skill for autonomous UI/UX orchestration.
• ziggygrep • Pure Zig grep replacement beating ripgrep 22 of 22 benches.
• jev-curate • High-throughput synthetic dataset sifter powered by TypeSafe AI.
• Portfolio • Projects, benchmarks, and writing.

License

MIT © Akash Priyadarshi