Codex Reset Timer

AGENTS.md vs Skills vs MCP: What Goes Where?

Put durable rules, reusable procedures and external connections in the right Codex layer.

7 min
The short answer

Use AGENTS.md for standing project rules, Skills for repeatable procedures, and MCP for live external systems.

01

The decision behind the technique

Mixing all three concerns in one giant instruction file makes updates risky and every task more expensive. Separation keeps context smaller and ownership clearer.

Use AGENTS.md for standing project rules, Skills for repeatable procedures, and MCP for live external systems.

02

A practical step-by-step workflow

Use this sequence as a default, then tighten or expand it only when the evidence changes.

  • Write repository-wide invariants in AGENTS.md
  • Package multi-step repeatable work as a Skill
  • Connect ticketing, docs or telemetry through MCP
  • Link layers instead of duplicating their content
03

Failure boundary and finish line

Do not turn a one-off workaround into permanent global instructions.

Finish when the configuration is smaller, reproducible and still preserves every required safety boundary.

Frequently asked questions

What should I do first?+

Write repository-wide invariants in AGENTS.md

What is the main failure mode?+

Do not turn a one-off workaround into permanent global instructions.

OFFICIAL SOURCES

Verify against the source

OpenAI customization overviewOpenAI MCP guidance

This guide provides general information, not an OpenAI allowance promise or support decision. Your account and official pages govern actual usage, benefits and recovery times.