skills.rest
illogical
illogicalCOMMUNITY
Explore all skills by this author →
GitHub

filesystem-context

Store and retrieve context data on the filesystem for long-running tasks.

Updated Jan 19, 2026
Search Tags:#filesystem#context-engineering#agent-memory#tool-output#plan-persistence#scratch-pad#context-discovery
One-click install
npx skills add https://github.com/illogical/InstructionLibrary --skill filesystem-context-illogical
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: filesystem-context
Source: https://github.com/illogical/InstructionLibrary/tree/main/skills/filesystem-context
Command: npx skills add https://github.com/illogical/InstructionLibrary --skill filesystem-context-illogical

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

The filesystem-based context pattern dramatically reduces prompt window pollution by persisting large tool outputs, logs, and intermediate results to the disk and loading only relevant references on demand. This enables dynamic context discovery for long-running tasks and multi-turn workflows without overwhelming the model.

Core Features & Use Cases

  • Scratch pad for tool outputs: offload large results to files and reference them in prompts.
  • Plan persistence: save long-horizon plans to disk and re-read them after context refresh.
  • Dynamic context discovery: selectively load context via file paths to minimize static prompt growth.
  • Sub-agent/workspace coordination: share findings and state via the filesystem across agents.
  • Dynamic skill loading: keep only skill references in memory and load content when needed.
  • Terminal/log persistence: persist terminal outputs for targeted retrieval and auditing.

Quick Start

To start using this skill, create a scratch folder, offload a large tool output and reference it in your next prompt, then save and load a plan from scratch/current_plan.json to resume work after context refresh.

Frequently Asked Questions about filesystem-context

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I save agent memory and reduce prompt window pollution during long-running workflows?▼

Filesystem context management reduces prompt window pollution by persisting large tool outputs, logs, and intermediate results to disk and loading only relevant references on demand. This enables dynamic context discovery for long-running tasks without overwhelming the model.

What is the best way to persist long-horizon plans for agents across context refreshes?▼

Plan persistence saves long-horizon plans to disk and re-reads them after a context refresh. By storing your current plan in a file like scratch/current_plan.json, you can easily resume work without losing state.

How do I share state and findings across multiple agents in a multi-step workflow?▼

Sub-agent and workspace coordination shares findings and state via the filesystem across agents. By offloading context data to accessible file paths, multiple agents can retrieve and update shared information deterministically.

Can I use filesystem context for dynamic skill loading without external tools?▼

Filesystem context requires no external tools beyond the standard environment. You can keep only skill references in memory and load content dynamically when needed, applying deterministic file-based patterns for accessible retrieval.

Does offloading tool outputs to a scratch pad work for terminal log persistence?▼

Terminal log persistence saves terminal outputs to disk for targeted retrieval and auditing. You can offload large results to a scratch folder and reference the file paths in your next prompt to minimize static prompt growth.

Related Skills

Explore More Software Engineering Skills→
closedloop-aiclosedloop-ai
102

closedloop-env

Provide ClosedLoop environment variables and construct file paths for agents.

Official
Basic
samvanmesamvanme

filesystem-context

Offload large tool outputs and persistent context to files using Python and JSON/YAML.

Community
Basic
eanzhaoeanzhao
1

mcp-filesystem-browse

Browse local workspace files using MCP filesystem tools.

Community
Basic
marinvchmarinvch

filesystem-context

Offload tool outputs and context to the filesystem for persistent agent workflows.

Community
Advanced