메뉴
Manage what goes into the AI agent context window for maximum quality and minimum waste
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Build UIs that work for all users including keyboard navigation, screen readers, and WCAG 2.2
Design multi-agent systems with robust tool interfaces, state management, and failure handling
Build ML systems with disciplined training, evaluation, deployment, and safety practices
Design APIs that are stable, ergonomic, and evolvable
Design systems at the right scale with explicit trade-off documentation
Design services that are reliable, observable, secure, and maintainable
| name | context-engineering |
| description | Manage what goes into the AI agent context window for maximum quality and minimum waste |
| difficulty | senior |
| domains | ["ai-ml"] |
Context is finite. What you put in the context window determines what the agent can reason about. Too much noise → the relevant signal is diluted. Too little context → the agent makes uninformed decisions. This skill manages context deliberately.
For your model and use case: how many tokens is your budget? Reserve: 20% for the system prompt, 20% for the output, 60% for the dynamic context (documents, history, tools).
Include in this order:
Cut: long documents that contain 1 relevant paragraph, full file contents when only a function is needed, conversation history beyond the last N relevant turns.
Agents pay more attention to the beginning and end of context. Put instructions at the top. Put the most relevant context closest to the task.
Mark different sections clearly:
<system>Your role and constraints</system>
<documents>Retrieved content</documents>
<task>What to do</task>
This prevents the model from confusing instructions with retrieved data.
Summarize long histories. Extract the relevant portions of long documents. Use structured data (JSON, tables) instead of prose where possible.
Track: output quality vs. context length. If longer context is producing worse results, you have a context quality problem, not a context quantity problem.