AGENTS.md directory

Astro

The web framework for content-driven websites.

Opens with four sections on how to approach work, before any repository detail.

Documents it routes to

This file points the agent at 3 other documents in the repository. Open any of them to read it here, as it was at this commit.


├── .agents
│   └── skills
│       └── writing-comments
│           └── 
└── reference
    ├── 
    └── 

By the numbers

165
Lines
1,310
Words
about 6 min read
8.7kB
File size
15
Headings
54
Bullet rules
2
Code blocks
4
Links to other docs
62.5k
Repo stars

File measured on main at commit 5be09d6. Analysis written 2026-09-14. Star count is a snapshot from September 2026.

What kind of file this is

The first third contains no Astro-specific content: it covers stating assumptions, scope, and success criteria. The rest is a tooling guide covering the monorepo layout, test invocation, background dev servers and browser automation.

Techniques in this file

Quoted lines are verbatim from the file. Open one to see it where it sits.

Four framing sections before any repository detail

Behavioral framing

Think Before Coding, Simplicity First, Surgical Changes, Goal-Driven Execution. None reference Astro. They address stating assumptions out loud, building only what was asked, leaving adjacent code alone, and defining what done means before starting.

A one-line test for scope

The Surgical Changes section ends with a test the agent can apply to its own diff line by line, rather than a disposition to hold.

A deletion test for comments

House vocabulary

A comment must state something the reader cannot recover from the code, written for someone reading at HEAD months later with no access to the conversation or the diff. Narrating change history and addressing the reviewer are both listed as prohibited.

Tasks rewritten as verifiable loops

Three conversions are given: "fix the bug" becomes "write a test that reproduces it, then make it pass". The stated reason is that criteria the agent can check let it iterate without returning to ask.

An instruction to fail rather than substitute

The browser automation section covers the case where the tool is missing and no human is available to install it. It directs the agent to fail the job rather than work around the absence.

A managed lifecycle for long-running servers

Dev servers are started through a four-command lifecycle, start, logs, status and stop, with a force flag for replacing a stale one. Backgrounding a server with an ampersand is listed as the thing not to do.

Takeaways for your own repo

  • The four opening sections contain no repository-specific content and transfer to another repo unchanged.
  • Give scope creep a one-line test the agent can run against its own diff.
  • Adopt the deletion test for comments. It addresses the "this now correctly handles" habit in one rule.
  • If your repo has long-running processes, document a start, logs, status, stop lifecycle rather than leaving it to the agent.

How the file is organized

  1. 01Think Before Coding
  2. 02Simplicity First
  3. 03Surgical Changes
  4. 04Goal-Driven Execution
  5. 05Style Guide
  6. 06Writing Comments
  7. 07Monorepo Structure
  8. 08Running Tests
  9. 09Background Dev Servers
  10. 10Deep Dives
Read the full file on GitHub

Context your AGENTS.md cannot carry

Astro's file tells an agent how the codebase works. It cannot tell it which bug three customers hit this week. Modem keeps that context current and attaches it to the work.