How does this prevent context window bloat?

By writing large outputs (like logs or search results) to files instead of returning them to the chat, the conversation history stays lean and focused on reasoning.

When should I activate the filesystem-context skill?

Activate it when tool outputs exceed 2000 tokens, tasks span multiple turns, or when you need to persist project plans beyond a single session.

What is dynamic context discovery?

It is a pattern where the agent receives minimal static pointers and uses search tools to load full content from the filesystem only when specifically needed for the task.

Can I use this for multi-agent workflows?

Yes, it allows different agents to share state and findings through a common file structure, reducing the need for verbose message passing.

Filesystem Context Engineering

Name: Filesystem Context Engineering
Author: monmacllcapp

bymonmacllcapp

0•

Productivity & Workflow

Manages agent memory and context bloat by offloading large datasets, plans, and tool outputs to the filesystem for just-in-time retrieval.

Filesystem Context Engineering provides a robust framework for managing effectively unlimited context during complex AI agent tasks. By treating the filesystem as a dynamic memory layer, this skill allows Claude to offload large tool outputs, persist long-term project plans, and facilitate communication between sub-agents without saturating the context window. It leverages standard filesystem tools like grep and glob to enable discovery-based context loading, ensuring the model remains focused on relevant information while maintaining access to deep project history and technical logs.

Key Features

01Dynamic context discovery for token-efficient information retrieval

02Persistent scratch pads for large tool outputs and database results

03Multi-agent state sharing through a localized filesystem workspace

04On-demand skill loading to minimize system prompt overhead

050 GitHub stars

06Terminal and log persistence for targeted error analysis

Use Cases

01Building self-correcting agents that store learned patterns in persistent files

02Executing long-horizon development tasks that require persistent plan tracking

03Handling massive web search or API results without bloating message history

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add monmacllcapp/skill-forks filesystem-context

For use in Claude.ai and ChatGPT

Download Skill