multi-layer-context-persistence

仓库创建 2026年7月6日最近提交 21 天前SkillHot 收录 21 天前

▸ 精选理由

解决 agent 冷启动与上下文丢失的实战方案。

这个 Skill 做什么

提供三层（Memory/Skills/会话检索）上下文持久化框架防止丢失。

提供三层上下文持久化：Memory（自动注入事实）、Skills（可复用工作流）和 Session Search（FTS5 检索），避免会话冷启动或丢失状态。当你上线新 agent、审计上下文管理或遇到“Dumb Zone”式的记忆丢失时用它。特点是多层协作、持续同步，不依赖单一手工交接文件。

▸ 展开 SKILL.md 英文原文

3-layer context persistence architecture — Memory (auto-inject facts), Skills (reusable workflows), Session Search (FTS5 retrieval). Load when onboarding new agents, auditing context management, or debugging "Dumb Zone" / context loss issues.

Skill 开发管理上下文持久化会话记忆检索层通用

Stars

Forks

仓库内 Skill

7 日增星

安装 / 使用

给你的 Agent 一句话（通用）

帮我安装这个 skill：https://raw.githubusercontent.com/lekalsutee/hermes-context-persistence/main/skills/multi-layer-context-persistence/SKILL.md

或 curl 直取 SKILL.md

curl -fsSL "https://raw.githubusercontent.com/lekalsutee/hermes-context-persistence/main/skills/multi-layer-context-persistence/SKILL.md"

SKILL.MD 节选查看完整文件 ↗

# Multi-Layer Context Persistence

## Overview

**One-shot handoffs (`handoff.md`) are a start.** They keep the next session from cold-starting. But they're manual, flat, and fragile — a single file you must remember to write, re-read every session, and manually keep updated.

**Our approach:** 3 independent persistence layers that work together continuously. The agent never loses context, never carries stale history, and learns from every interaction.

> The "Dumb Zone" (150K-200K tokens of stale context) doesn't happen because active context stays at 5-15K tokens. Memory and skills are injected/loaded on demand; history is searchable but not present.

---

## When to Use

- **Onboarding a

via SKILL·HOT · 数据来自 GitHub 公开信息 · 原文版权归作者所有