AgentBase

AgentBase is built by Chris Izworski — emergency services technologist, former Michigan 911 Executive Director, and someone who has spent a lot of time watching AI agents fail because their instructions were bad.

The web was not built for agents. Sites are full of JavaScript walls, inconsistent formatting, pagination traps, and content that assumes a human is doing the reading. Agents trying to retrieve instruction files hit all of it. There was no clean, structured place to find and serve agent instructions. So this is that place.

It is a start. The catalog is small. The infrastructure is intentionally simple — flat files, clean endpoints, no authentication required. That is by design. Agents shouldn't need to jump through hoops to get what they need.

Editorial Standards

Not everything submitted gets published. Every file is reviewed against these principles before it goes live:

• 01 It has to actually workLoad the instruction into an agent and it should do what the file says it does. Vague, untested, or placeholder instructions are rejected.
• 02 It cannot facilitate harmInstructions that help agents deceive, manipulate, harm, or exploit people are not published here. This is enforced with automated scanning and manual review.
• 03 It has to be legalSubmissions must be original work or properly licensed. Instructions that facilitate illegal activity are rejected. If you built it, you can submit it.
• 04 Quality scores are honestA 70 is a 70. A 95 is a 95. Featured picks are my call and I stand behind them. The scoring rubric is public — you can see exactly how scores are calculated.

Where This Goes

Phase 1 is a free registry. Files are CC0 or permissively licensed. Anyone can fetch them. The infrastructure is agent-native from day one — /registry/index.json, raw markdown endpoints, /llms.txt.

Eventually: paid listings, micro-transactions for premium instruction sets, and direct agent submission when the trust infrastructure is in place. The model that makes sense for the agentic web — whatever that turns out to be — will get built here as it becomes clear what agents actually need.

The goal is a registry agents can trust. Instructions that are accurate, versioned, rated, and safe. That takes time to build. We're building it.