What are you md files best practices?

Codex
1

I’ve noticed just in the last week that 5.5 Extra High has started recording bits and pieces into the file ~/.codex/TASK.md and I’m wondering if there’s a best practice for telling it which md files to create and use? I’ve also had it just keep track of things it discovers in a README.md in my project while trying to optimize some webgl2.

How is everyone else helping it keep track of important data over a long goal?

2

If you had asked that several months ago, the answer would have been simpler. At that point, there were only a few Markdown files with fairly specific names and roles that were commonly used to shape the prompt and project context sent to the model.

Unrolling the Codex agent loop

Today, the answer is more dynamic. Codex can still be guided by project documentation such as AGENTS.md, which OpenAI describes as a file for telling Codex how to navigate the codebase, what commands to run, and what project practices to follow. However, with newer workflows such as the /goal slash command and codex exec for non-interactive/scripted runs, the set of Markdown files involved may change during the task.

In practice, I would not think of this as “which fixed Markdown files should I create?” so much as:

  1. Put stable project-level instructions in AGENTS.md.
  2. Use README.md or other normal project docs for human-readable context.
  3. Let task-specific Markdown files be created, updated, or removed as needed during the run, especially for longer /goal or codex exec workflows.
  4. Keep important instructions and durable project knowledge in predictable files, rather than relying on temporary notes the agent may generate while working.