filesystem-context

Offload agent context to filesystem files for dynamic retrieval and persistence.

Updated Apr 7, 2026
One-click install
npx skills add https://github.com/NT-boop-star/BRMV-tract --skill filesystem-context-nt-boop-star
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: filesystem-context
Source: https://github.com/NT-boop-star/BRMV-tract/tree/main/antigravity/skills/filesystem-context
Command: npx skills add https://github.com/NT-boop-star/BRMV-tract --skill filesystem-context-nt-boop-star

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve? LLM context windows are limited, and large tool outputs, long-running plans, and multi-agent communication quickly bloat or overflow the window. This Skill uses the filesystem as an overflow layer so agents store, retrieve, and update effectively unlimited context through files instead of message history. ## Core Features & Use Cases - Tool Output Offloading: Automatically write large tool outputs (over ~2000 tokens) to scratch files and return compact summaries with file references. - Plan Persistence: Save structured multi-step plans to disk so agents can re-read objectives and progress after context refreshes or summarization. - Sub-Agent Workspaces: Route sub-agent findings through isolated per-agent directories instead of lossy message-passing chains. - Use Case: A web search returns 8000 tokens of results. The Skill writes them to scratch/search_results_001.txt, returns a 100-token summary with the file path, and the agent greps the file later only when specific details are needed. ## Quick Start Ask the agent to offload large tool outputs to scratch files and persist the current task plan to disk so it survives context window refreshes.

Frequently Asked Questions about filesystem-context

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

FAQPage Schema
How do I reduce LLM context window bloat from tool outputs?

Write tool outputs exceeding roughly 2000 tokens to scratch files and return only a compact summary with the file path. The agent then uses grep or line-range reads to retrieve specific details on demand instead of keeping the full output in message history.

How do agents persist plans across context window refreshes?

Serialize the plan as structured JSON or YAML with objective, step statuses, and notes, then save it to a file like scratch/current_plan.json. The agent re-reads this file at the start of each turn or after summarization to restore task awareness.

When should I use filesystem-based context management?

Use it when tool outputs exceed about 2000 tokens, tasks span multiple turns, or multiple agents share state. Avoid it for single-turn tasks, when context fits comfortably in the window, or when file I/O latency is unacceptable.

How do sub-agents share findings without message passing?

Each sub-agent writes findings to its own isolated workspace directory, and the coordinator reads those files directly. This preserves full fidelity and avoids the information degradation that occurs in multi-hop message chains.

What are the risks of agent self-modification of preference files?

Self-modification can accumulate incorrect or contradictory instructions over time. Guard it with validation such as entry limits, value length caps, and periodic human review of persisted preferences.

Why do stale file references break filesystem context workflows?

Agents may hold paths from prior turns that no longer exist after refactors or file moves. Always verify file existence before reading a cached path and re-discover files with glob if the check fails.