tiered-memory

Organizes agent memory into hot, cold, and wiki tiers for context reduction per spawn.

2|Updated Jul 24, 2026
One-click install
npx skills add https://github.com/elbruno/ElBruno.MagenticUI --skill tiered-memory-elbruno
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: tiered-memory
Source: https://github.com/elbruno/ElBruno.MagenticUI/tree/main/.squad/templates/skills/tiered-memory
Command: npx skills add https://github.com/elbruno/ElBruno.MagenticUI --skill tiered-memory-elbruno

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Squad agents currently load their full context history on every spawn, which grows unboundedly across sessions and wastes context window space. This Skill defines a three-tier memory model so agents only load the bytes relevant to the current task. ## Core Features & Use Cases - Three-Tier Memory Model: Separates memory into Hot (current session, always loaded), Cold (summarized cross-session history, loaded on demand), and Wiki (durable structured knowledge, selectively read). - Spawn Template Pattern: Provides a ready-to-use spawn prompt template that defaults to Hot-only context, with optional --include-cold and --include-wiki flags for history or domain knowledge. - Scribe Integration Design: Describes how a Scribe agent compresses Hot into Cold at session end and promotes aged Cold entries into Wiki knowledge. - Use Case: When resuming interrupted work, spawn an agent with Hot context plus the Cold summary file instead of replaying full transcripts, cutting context payload size significantly. ## Quick Start Use the tiered-memory spawn template to launch an agent with only hot session context, adding the cold or wiki sections when the task needs history or domain knowledge.

Frequently Asked Questions about tiered-memory

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

FAQPage Schema
How do I reduce agent context size when spawning AI agents?

Use a tiered memory model that loads only hot session context by default instead of full history. Add cold summarized history or wiki reference docs on demand via include flags when the task actually needs them.

What is the difference between hot, cold, and wiki memory tiers?

Hot holds current session context (~2-4KB) and is always loaded. Cold stores summarized cross-session history loaded on demand. Wiki contains durable structured knowledge like ADRs and conventions, read selectively.

Is the tiered memory runtime implemented in Squad?

No, this is a design proposal only. The tier scaffolding, Scribe promotion logic, and tier-aware spawn reads are tracked in bradygaster/squad#1264. Until implemented, agents continue loading full history.md and decisions.md on every spawn.

When should I include cold memory in a spawn template?

Include cold memory when resuming interrupted work, debugging a recurring issue, or answering what was tried before. Pass the --include-cold flag and fetch the agent's cold summary file before spawning.

How does the Scribe agent manage memory tiers?

Per the design, Scribe compresses hot context into a cold summary at session end (targeting about 10% of session verbosity) and promotes aged cold entries into wiki docs when they become stable knowledge. This runtime behavior is not yet implemented.