360 Secure SkillHub
360 安全技能中心

mem-load

该技能定义了记忆加载流程:先读取 activeContext.md 与 session_cache.md,再据此识别当前任务与最近会话,按需加载任务文件、会话文件和实现细节文档,从而在开始工作前恢复项目上下文。仅描述读取逻辑,不含写入或更新记忆的能力。

space-cadet
space-cadet
热度 14.60
最近更新 2026-08-21 16:02:35
分类:知识依赖:读写文件应用场景:工作流 / 笔记整理 / 任务管理

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add mem-load
Metadata
name
mem-load
description
Global workflow for loading memory bank files based on recent activity and context Use when the user invokes $mem-load or asks for this workflow by name.

mem-load

This skill is a Codex conversion of a personal Windsurf global workflow. Follow the workflow below, adapting Windsurf-specific slash-command wording to Codex skill invocation. If the workflow mentions running /mem-load, treat that as explicit invocation of $mem-load.

Memory Load Workflow (mem-load)

Purpose

Load relevant memory bank files by analyzing recent activity and context to determine which implementation docs are needed.

Loading Logic

Step 1: Load Core Context Files

read memory-bank/activeContext.md
read memory-bank/session_cache.md

Step 2: Identify Most Recent Activity

Analyze the loaded context to determine:

  • Current Task: Which task ID is marked as active focus
  • Recent Session: Which session file has the most recent timestamp
  • Last Updated: Which task or session files were modified most recently

Step 3: Load Recent Task/Session Files

Based on the analysis from Step 2:

# Load current task file if ID identified
read memory-bank/tasks/T{ID}.md

# Load most recent session file
read memory-bank/sessions/YYYY-MM-DD-PERIOD.md

Step 4: Determine Implementation Docs to Load

Based on the task and session context:

Load these implementation docs if referenced or relevant:

  • Any implementation-detail files mentioned in task file
  • Any files referenced in session working state
  • Technical context if implementation approach unclear
# Load implementation docs as needed (examples)
read memory-bank/implementation-details/[specific-file].md
read memory-bank/techContext.md  # Only if technical details needed
read memory-bank/systemPatterns.md  # Only if architectural context needed

Decision Logic

START
├─ Load activeContext.md + session_cache.md
├─ Identify current task ID and recent session file
├─ Load current task file + recent session file
├─ Check for implementation doc references
│  ├─ References found → Load those specific docs
│  └─ No references → Load based on task type
│     ├─ Implementation task → Load techContext.md
│     ├─ Architecture task → Load systemPatterns.md
│     └─ Other task → No additional docs needed
└─ Complete loading

Loading Commands

Core Load (Always)

read memory-bank/activeContext.md
read memory-bank/session_cache.md

Contextual Load (Based on analysis)

# Task file (if current task identified)
read memory-bank/tasks/T{ID}.md

# Recent session file
read memory-bank/sessions/YYYY-MM-DD-PERIOD.md

# Implementation docs (only if referenced or needed)
read memory-bank/implementation-details/[specific-file].md
read memory-bank/techContext.md
read memory-bank/systemPatterns.md