Instruction File Length Question Explained
This worked Harness Engineering practice question tests Maps and Instructions. Try the answer first, then review why C is the best choice.
Short answer
The correct answer is C. Cut it to a short map that names the file holding each detail.
OpenAI's Codex team says to "give Codex a map, not a 1,000-page instruction manual." A giant instruction file crowds out the task and the code, and when every rule is marked important, no rule stands out. The team keeps AGENTS.md to roughly 100 lines, as a table of contents that points to deeper files. Claude Code's docs give the same advice for CLAUDE.md: target under 200 lines, because longer files consume more context and reduce adherence. The correct answer is C. Cut it to a short map that names the file holding each detail
The Question
Your AGENTS.md has grown to 900 lines. The agent now ignores rules that are clearly written in it. What should you change?
Why C is correct
OpenAI's Codex team says to "give Codex a map, not a 1,000-page instruction manual." A giant instruction file crowds out the task and the code, and when every rule is marked important, no rule stands out. The team keeps AGENTS.md to roughly 100 lines, as a table of contents that points to deeper files. Claude Code's docs give the same advice for CLAUDE.md: target under 200 lines, because longer files consume more context and reduce adherence. The correct answer is C. Cut it to a short map that names the file holding each detail
Why the other options are wrong
If every rule is marked important, the marker carries no signal. The problem is the amount of text that competes for context, and the labels add more text.
Repeated rules make the file longer, and length is the cause of the problem. The agent still has to find each rule among hundreds of others.
Two files that both load at start cost the same context as one 900-line file. Claude Code's docs say imported files load together with the file that imports them.
Key idea: Maps and Instructions
Every question here is a situation you meet with Claude Code or Codex: the agent gets something wrong, and you choose what to change. The best answer is usually a file, a check, or a hook that stops the mistake from repeating, not a louder prompt. If you miss a question, read the explanation and note which part of the harness it uses: the AGENTS.md map, a skill, a check, a visitor check, a hook, or a handoff note.
Ready to test the full skill?
Take the free Harness Engineering quiz and find the exact basics to review next. No signup required.
Practice related questions
Same skill area or adjacent concept. Use these after reviewing the explanation.