context-hygiene

Enforce context budgets and delegate bounded tasks to subagents.

2|Updated Jun 6, 2026
One-click install
npx skills add https://github.com/lotea-be/qrspi --skill context-hygiene-lotea-be
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: context-hygiene
Source: https://github.com/lotea-be/qrspi/tree/main/claude/skills/context-hygiene
Command: npx skills add https://github.com/lotea-be/qrspi --skill context-hygiene-lotea-be

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Keeps AI sessions focused by enforcing lean context windows and using subagents as bounded work units rather than persistent personas. Load this when planning a multi-stage workflow, starting a long session, or deciding whether to delegate work to a subagent.

Core Features & Use Cases

  • Enforces a target context window usage (<40%) and hard reset at 60%.
  • Treats subagents as context firewalls, isolating duties and preventing prompt leakage.
  • Provides orchestration guidance, when to persist progress to disk, and how to delegate bounded tasks to subagents.
  • Use Case: Orchestrate a multi-stage research flow with bounded artifacts to avoid context bloat.

Quick Start

Load this skill at the start of a long, multi-stage session to enforce context budgeting and subagent-based task delegation.

Frequently Asked Questions about context-hygiene

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

FAQPage Schema
How do I prevent context bloat in long-running AI agent sessions?

Prevent context bloat by enforcing lean context windows and using subagents as context firewalls to isolate bounded tasks. This maintains a target context usage below 40% and hard resets at 60% to keep multi-stage workflows focused.

How do I use subagents for task delegation without leaking sensitive prompts?

Use subagents for task delegation without leaking sensitive prompts by treating them as context firewalls. This isolates duties and routes bounded work units safely, preventing persistent persona memory from accumulating or exposing sensitive data.

When should I reset context during a multi-stage AI workflow?

Reset context during a multi-stage AI workflow when usage hits a 60% hard limit. Orchestration guidance dictates persisting progress to disk before resetting, ensuring extended dialogues and bounded artifact writes continue without losing state.

Does this context management approach work for multi-stage research flows?

This context management approach works for multi-stage research flows by enforcing clear context budgets and delegating bounded tasks to subagents. It orchestrates extended dialogues and writes bounded artifacts to avoid context bloat.

What is the best way to orchestrate subagents safely in extended dialogues?

The best way to orchestrate subagents safely in extended dialogues is enforcing controlled delegation with context firewalls. This prevents prompt leakage and isolates duties, ensuring safe orchestration without exceeding context budgets.