AgentReady

"Battle-test your AI agent's knowledge before you ship. Ship with data, not vibes."

AgentReady is a lightweight, Netlify-deployed harness that stress-tests the knowledge base and agent harness (skills/rules/tools/system prompts) you built around an LLM.

What it does

1. Upload your agent's knowledge base (docs, runbooks, READMEs, etc.)
2. Upload your agent harness (optional) + describe the agent's mission
3. Get a readiness score and a prioritized list of blind spots before users find them

How it works

1. Generates stress-test probes tailored to your mission and domain (incidents, edge cases, adversarial prompts, cross-domain questions)
2. Evaluates knowledge + harness coverage by running probes and forcing the model to be explicit about confidence + missing context
3. Returns a readiness score + top gaps (what's missing, whether it's a knowledge vs harness issue, why it matters, and how to fix it)

Tech stack

• Netlify Functions (serverless)
• Anthropic Claude API (@anthropic-ai/sdk)
• Vanilla JS (no framework)
• Neo-brutalist CSS

Setup

Local dev

git clone https://github.com/rajnavakoti/agent-ready
cd agent-ready
npm install

# Run Netlify dev server (functions + static site)
ANTHROPIC_API_KEY=your-key npm run dev

Then open the local URL printed by Netlify (defaults to http://localhost:8888).

Deploy to Netlify

1. Create a new Netlify site from this repo
2. Set the environment variable ANTHROPIC_API_KEY
3. Deploy

netlify.toml already sets:

• publish = "src"
• functions = "netlify/functions"

Architecture

The app is a 3-function pipeline:

• netlify/functions/generate-probes.mjs: given mission + knowledge base + harness, generates targeted probes (JSON)
• netlify/functions/execute-probes.mjs: runs each probe against the provided context and returns answers + confidence + explicit gaps
• netlify/functions/analyze-gaps.mjs: aggregates results into category scores, an overall readiness score, and the top blind spots

On the frontend, src/js/app.js orchestrates the pipeline:

1. /.netlify/functions/generate-probes
2. /.netlify/functions/execute-probes
3. /.netlify/functions/analyze-gaps

Datasets

AgentReady uses 17+ public datasets across 8 domains to ground scenario simulation, including:

• Customer support
• IT incidents
• Healthcare
• Legal
• Finance / fraud
• E-commerce
• Real estate
• Code / bugs

Research

This project is inspired by DDC-style evaluation methodology: https://arxiv.org/abs/2603.14057

License

MIT (see LICENSE).