How AgentMD Works

A plain-language explanation of what AgentMD does and how it works.

What AgentMD Does (Simple Version)

AGENTS.md is like a recipe for AI coding tools. It says things like "run tests," "run the build," "run the linter." Most teams only read it. AgentMD actually runs those steps and checks that they work.

The Three Main Parts

1. Read & Understand (Parse)

AgentMD reads your AGENTS.md and figures out: what sections it has (Build, Test, Lint, etc.), what commands to run (e.g. pnpm test, pnpm build), and any rules or limits you've set (e.g. "don't run dangerous commands").

Think of it like reading a recipe and listing all the steps before you start cooking.

2. Check It's OK (Validate)

Before running anything, AgentMD checks: the file isn't empty, there are no obviously dangerous commands (e.g. "delete everything"), and it has the usual sections (build, test, etc.).

It also gives a score from 0–100 for how "ready" your AGENTS.md is for AI tools.

3. Actually Run It (Execute)

AgentMD runs the commands from your AGENTS.md in order: parse the file, decide which commands to run, run them in a safe way (with time limits and safety checks), and record what happened (passed, failed, how long it took).

Why It Matters

Without AgentMD, AI tools might skip steps or do things differently each time. With AgentMD, the same steps run every time, in the same order, and you get a clear record of what ran and whether it passed.