About Agent-Ex

Agent-Ex started as a research question: what can you actually customize across agentic coding tools, and how do the options compare? The initial research was conducted using ChatGPT's Deep Research mode, which crawled official vendor documentation and produced a detailed comparison report cataloging every documented extension point, its scope, interfaces, availability, trust model, and upstream sources.

That report was thorough but hard to navigate as a wall of text. Agent-Ex turns it into an interactive web app where you can quickly scan a comparison matrix, drill into any tool or extension family, and follow source links to verify facts against the upstream docs. The goal is to track extension point capabilities across whatever vendors and families the research turns up over time — not just a fixed set of tools.

How it's built

The app is built with SvelteKit and Tailwind CSS, deployed to Cloudflare Pages. The data lives in simple Markdown files with YAML frontmatter — one file per vendor in the data/ directory. This means humans or agents can update the data without touching any application code. New vendors and new extension families can be added by editing data files alone.

A skill update-vendor-data/ provides instructions for agent-assisted updates. An agent can search vendor documentation domains, fetch upstream URLs, compare against current data, and propose edits — including updating the skill itself with newly discovered source URLs.

Normalized taxonomy

To make cross-tool comparison meaningful, vendor-specific terminology is normalized into extension families grounded in what users recognize across tools:

• Instructions — persistent context the agent sees every session
• Prompts — reusable prompt templates explicitly invoked by name
• Skills — richer workflows and knowledge the agent can discover and use on its own
• MCP & Tools — connections to external services via Model Context Protocol
• Agents — specialist personas, isolated workers, or coordinated teams
• Hooks — deterministic automation on lifecycle events
• Plugins & Distribution — packaging and sharing bundles of the above
• Settings & Policy — tool configuration, managed settings, and organizational controls

This taxonomy evolves as vendors add new concepts. If a new extension type doesn't fit an existing family, a new one can be added to data/_meta.yaml.

Source and accuracy

Every data point links to its upstream documentation source. Since vendor docs change frequently, some information may be outdated. Use the source links to verify current status, and consider running the update skill to refresh the data.