Context Engineering

You might say

As the task gets longer, the AI starts missing a newly confirmed requirement. What should the next turn include?

Organizing the goals, rules, source material, and latest results an AI receives for each turnFor a long login fix, keep the current error, relevant files, and the rule that shared components cannot change. Repeated logs can become a summary with a path to the original.

Know first

Context Window System Prompt Conversation History

Context Management

What should a long task keep for its next turn?

Available material

Context for this request · budget 1000 Token

Goal for this turn120 Token

120 / 1000 · reserved for the answer 880

Keep the task for this turn first so the Agent can judge which other material is relevant.

Send only what the next turn needs; preserve older clues with a summary and file location.

How to preserve requirements without resending every record

Keep current requirements and rule carriers visible: Carry the goal, protected scope, and confirmed decisions into the next turn; project rules files (like AGENTS.md, .cursorrules) preserve conventions across separate chats.

Replace repetition with a summary: Keep new decisions and open questions, but reduce repeated chat and old logs to conclusions that still matter.

Read the original only when needed: Large originals do not need to be resent every turn, but their paths and sources must remain available.

Quick checkChoose the best answer

A long task contains requirements, approved architecture decisions, dozens of pages of old logs, and repeated tool output. The agent starts missing the newest constraint. How should the next turn be prepared?

Insert the entire history and every raw tool result again so no word can be omittedMake only the final instruction more forceful and leave all history and tool results unchangedKeep the goal, constraints, architecture decisions, and open issues; compress old logs and retrieve originals by path only when needed

You can say this to an AI Agent

This task is long, and you have started missing a newly confirmed constraint. Before continuing, prepare a working context with the current goal, protected scope, approved decisions, relevant files, and open issues. List repeated logs and obsolete attempts as material to exclude, while keeping paths for on-demand retrieval. Show me the result for confirmation before proceeding.

Learn next

1Model Context Protocol→2Skill→

Want GPT or Codex?VibeHubGPT support

GPT-6 Astra

Build in 3D. Put your computer to work.