Can multiple agents use this skill simultaneously?

Yes, the filesystem acts as a shared memory layer, allowing different agents or sub-agents to communicate and share state without direct message passing.

What is filesystem-based context engineering?

It is a technique where an AI agent uses the local filesystem to store and retrieve data on-demand, rather than keeping everything in its active context window.

When should I use the scratch pad pattern?

Use it whenever a tool returns more than 2,000 tokens of data that isn't immediately critical for the current turn but may be needed for reference later.

How does this skill reduce token costs?

By writing large tool outputs to files and only reading summaries or specific lines, it prevents the conversation history from growing unnecessarily large.

How does the agent find information stored in files?

The agent uses standard filesystem tools like ls, glob, and grep to discover file structures and search for specific patterns or data points dynamically.

Filesystem-Based Context Engineering

Name: Filesystem-Based Context Engineering
Author: muratcankoylan

bymuratcankoylan

•

7,304

生産性とワークフロー

Optimizes agent context management by offloading large data, tool outputs, and persistent state to the filesystem for on-demand retrieval.

概要

The Filesystem-Based Context Engineering skill addresses the fundamental limitation of context windows by treating the local filesystem as a dynamic, scalable memory layer. By implementing patterns like scratch pads for large tool outputs, persistent plan tracking, and cross-agent communication files, this skill allows Claude to handle complex, long-horizon tasks without bloating its message history. It enables agents to selectively load information using search tools like grep and line-specific reads, significantly reducing token consumption while maintaining access to critical data and learned preferences.

主な機能

Enables asynchronous state sharing between multiple sub-agents
7,304 GitHub stars
Offloads large tool outputs to scratch files to prevent context window bloat
Facilitates selective terminal log analysis through targeted grep operations
Persists execution plans to maintain focus during long-horizon tasks
Supports dynamic skill loading to keep system prompts lean and efficient

ユースケース

Processing massive web search or database results by storing raw data and returning summaries
Building multi-agent architectures that require a shared 'source of truth' for context
Maintaining task continuity across multi-turn sessions by reading persistent plan files

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add muratcankoylan/agent-skills-for-context-engineering filesystem-context

For use in Claude.ai and ChatGPT

Download Skill

GitHub

概要