- ▸Listing the files
- ▸Reading the instructions
- ▸Checking the build tools
- ▸Finding the tests
- ▸Reading the folder layout
- ▸Adding up the score
vercel/next.js
Needs work
The overall score is the average of these six
Open a category to see every check
Can an agent quickly find the code that matters?
Biggest wins first
This project talks to a database but nothing documents how to change the schema. Agents commonly hand-edit generated migration files when this is undocumented.
Add a Database section to AGENTS.md describing:
746 routes were detected with no written convention for request validation, response shape, or error format.
Add an API section to AGENTS.md describing:
Not available yet
An AI reads the code to settle the 3 checks a quick scan cannot. Takes about a minute and is free with a GitHub sign-in.
Are agent-facing instructions present and complete?
Can an agent verify its own changes?
Does the codebase repeat the same patterns?
Is setup, build, lint and test behavior obvious?
How much must an agent read to make one change?
No template lists the environment variables this project needs, so an agent cannot tell which ones a feature depends on.
Commit a .env.example that lists every variable with a comment describing it.
The README is 4 words. It is the first file an agent reads.
Cover what the project does, how to run it, and where the main code lives.
Nine in ten files sit 9 levels deep. Deep paths are hard to guess, so agents fall back to listing directories.
Flatten folders that only contain one child folder.
Tests are stored away from the code they cover, so an agent must search to find the test for a file it just changed.
Move tests next to their source file, or document the mapping in AGENTS.md.