Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesAn effective AGENTS.md tells a coding agent project-specific facts it cannot safely infer from the code, then turns those facts into clear actions it can follow and checks you can verify. Put broadly useful conventions at the repository root, scope specialized rules to the directories or tasks they govern, and confirm both that your chosen agent loads the file and that it follows it on a representative task.
What should I put in an AGENTS.md file?
Use the file for guidance that is specific to your repository and would be difficult to infer reliably from its source code. OpenAI recommends recording items such as conventions, business logic, known quirks, and dependencies. The goal is not to repeat generic advice about writing good code; it is to supply context that changes how an agent should work in this project. OpenAI describes the file as a way to help Codex operate more effectively across prompts in its Codex best-practices guidance.
- Conventions: State repository-specific naming or implementation rules, with the relevant files or components in scope.
- Business logic: Explain domain rules an agent could misunderstand from a local code change alone.
- Known quirks: Describe constraints or non-obvious behavior that should affect edits.
- Dependencies: Identify important dependencies and any project-specific requirements for using them.
- Validation: When verified for this repository, name the commands or checks that establish whether a change is correct.
Do not assume an example architecture, directory, or command applies to your project. Inspect the repository and include only paths, procedures, and requirements that are accurate for it.
How do I write effective AGENTS.md instructions?
Write rules as observable actions. A useful instruction identifies where it applies, what the agent should do, and—where practical—how to check the result. “Keep database access in src/repositories” is more actionable than “use clean architecture,” but only use that path if it is actually part of your repository. Microsoft’s VS Code agent-customization guidance uses this kind of concrete convention to illustrate project instructions.
#1 Best Overall
For each rule, make its scope and activation conditions explicit. If it applies only to a language, module, file pattern, or test area, say so rather than making every task carry unrelated detail. Prefer a rule that a maintainer can check in a diff, command result, or other concrete output over one that depends on an agent’s interpretation of a broad aspiration.
OpenAI’s general agent-instructions guidance recommends smaller, clearer steps and explicit actions or outputs. In practice, split a broad requirement into focused directions: name the relevant area, specify the expected change, and state the check that confirms completion when one is known. Keep instructions concise enough to remain usable, but include the project context needed to avoid incorrect assumptions.
How should I organize root, nested, and targeted instructions?
| Location or mechanism | Best suited to | Trade-off to manage |
|---|---|---|
Repository-root AGENTS.md |
Conventions and context that are relevant across the repository. | Keep it focused: irrelevant rules burden tasks that do not need them. |
Nested AGENTS.md |
Guidance that genuinely applies to a smaller directory tree. | Check for conflicts with broader instructions and confirm the target harness discovers nested files as expected. |
| Harness-specific targeted instructions | Rules that apply only to selected files, languages, or task types, when the harness supports targeting. | Format and activation are tool-specific; maintain consistency with shared project requirements. |
Microsoft recommends matching instruction scope to intended use. In VS Code’s documented customization format, .instructions.md files can use applyTo patterns and descriptions; Claude rules use paths. These are harness-specific mechanisms, not a universal AGENTS.md syntax.
How do nested AGENTS.md files work in Codex?
Codex documentation describes an AGENTS.md as applying to the directory tree rooted at its location. For files touched in a patch, all applicable instructions must be followed. When instructions conflict, deeper files take precedence over broader ones; direct system, developer, and user instructions take precedence over AGENTS.md. The inspected Codex implementation comments describe finding the project root, collecting files along the path from that root through the working directory, and not traversing above the project root. These statements describe Codex’s documented behavior, not a rule every agent harness shares. See the Codex AGENTS.md documentation pointer and verify current version or configuration behavior when it matters.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallDoes AGENTS.md work with multiple AI coding agents?
It can serve as shared guidance when the tools you use support it, but the filename alone does not guarantee identical discovery, scope, precedence, or activation across harnesses. Microsoft explicitly cautions that behavior depends on the selected harness. Check each agent’s current documentation rather than assuming that one tool’s loading rules apply to another.
Where a tool requires its own native instruction file, keep its contents consistent with shared repository requirements and avoid contradictory copies. Use targeted features when they help a tool apply specialized guidance only to relevant files, while keeping cross-tool requirements aligned.
Rank #4
How can I tell whether my coding agent is following AGENTS.md?
Test discovery and behavior separately. A harness showing or listing an instruction file confirms that it found the file; it does not establish that the agent followed its rules. Microsoft’s VS Code guidance makes this distinction explicit.
- Check discovery in the target harness. Use its documented interface or status view to confirm that the intended instruction file is active for the task.
- Start a fresh conversation when appropriate. This avoids relying on earlier context that may have supplied the same rule independently.
- Choose a small, representative task. Give the agent a task for which one instruction has a clear, observable consequence.
- Define the success criterion before running the task. For example, check whether a change stays within the stated file scope or uses the required project-specific procedure.
- Inspect the result and tool activity. Compare the edits and actions with the rule; finding the file in a listing is not a substitute for this check.
If the result misses the rule, first confirm that the file’s scope covers the touched files and that the harness loaded it. Then make the instruction more explicit about its action, scope, or success check and repeat the representative task. Review generated instruction files before adopting them: Microsoft’s guidance warns that generated paths, commands, or conventions may be incomplete and should be checked against the repository.
Free tools Windows power users keep installed
One-click scans. No signup required.
Best Value
These recommendations reflect the cited OpenAI and Microsoft guidance available on October 3, 2026; live product documentation and implementations can change.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
