For AI agents: llms.txt is a short guide that sends you to the rules your problem needs, and rules.json holds every rule of this playbook as JSON.

How to do it

AI tools learn from public code, docs and answers. Software that is old, popular and stable fills their training data, so agents write it correctly and spot its mistakes. A new major version changes APIs after the training cutoff, so agents mix old and new syntax. Pin the major version and move up once it has been out for six months and your agent writes it without errors. Put the exact versions in the agent instructions file, so the agent does not guess.

Examples

Using these examples: they show one way to follow this rule. Keep the behaviour and the check, not the syntax. Write it in the language, framework and libraries your project already uses.

JavaScript example

// Pin the major version the agent knows, and say so in the instructions file
// package.json
// "dependencies": { "react": "^19.1.0" }
const pinnedMajor = range => /^\^(\d+)\./.exec(range)?.[1];

Python example

# Pin the major version the agent knows, and say so in the instructions file
# pyproject.toml: dependencies = ["django>=5.2,<6"]
import re

def pinned_major(spec: str) -> str | None:
    match = re.search(r"<(\d+)", spec)
    return str(int(match.group(1)) - 1) if match else None

How to check

Two different AI coding assistants each build the pick's official getting-started example with no documentation pasted in. Both results run and pass the example's own test on the first or second try.

Related rules

• Write one instructions file for agents in AI development
• Use React for the front end in Software choices
• Use OpenTelemetry to instrument your code in Software choices