Context Engineering

Context Engineering

Techniques for context engineering with AI agents — managing context windows, prompt structure, and knowledge organisation.

Claude / Claude Code
GitHub Copilot
Cursor
VS Code
OpenAI Codex
Gemini CLI
DevelopmentDeveloperResearcher

What it does

Long-running agent tasks fail because the context window fills with irrelevant information — previous steps, error messages, intermediate outputs — until the model loses track of the actual goal. Context engineering is the discipline of deciding what to keep, what to compress, what to summarise, and what to discard at each step of a multi-step workflow. This skill loads specific techniques for managing context in agent pipelines: windowing strategies, compression patterns, state externalisation, and the signals that tell you context is degrading before the task fails.

Use case

Building or debugging long-running agent workflows where Claude loses the plot midway through. Also useful for anyone working with large codebases where naively including all context causes performance to degrade. Made by muratcankoylan.

The Prompt

Copy and use immediately
"Compress this context down to what Claude needs to continue this task."
"Design a context management strategy for this 10-step research workflow."
"My agent is losing track of the goal after step 5 — diagnose the context problem."
"Externalise the state from this agent so the context window stays clean."
"Build a context window that keeps the goal pinned and rotates out completed steps."

How to use

  1. 1

    Describe the multi-step workflow and where it starts degrading.

  2. 2

    Claude diagnoses the context management issue and recommends a specific windowing or compression strategy.

  3. 3

    For new workflows: Claude designs the context architecture upfront, not as a retrofit.

Input / Output

Input

A description of the agent workflow, the steps it takes, and the point where context degradation becomes a problem.

Output

A context management strategy: what to keep in the active window, what to compress into summaries, what to externalise to files, and the checkpoints where context should be reviewed.

Added 15 Mar 2026Submitted by muratcankoylan👁 47📋 0

Details

Platforms
Claude / Claude CodeGitHub CopilotCursorVS CodeOpenAI CodexGemini CLI
Category
Development
License
apache-2.0

Stats

📋 Copies0
👁 Views47
👍 Upvotes0

Install with skills.sh

npx skillsadd muratcankoylan/skills/context-engineering

Requires skills.sh CLI

Community Notes

Sign in with GitHub to leave a note.

No notes yet. Be the first to contribute.