360 Secure SkillHub
360 安全技能中心

Dory-Proof Memory System

用工作区 Markdown 文件（AGENTS/SOUL/USER/MEMORY/LESSONS 及 state、memory、ops 目录）实现 AI Agent 跨会话记忆：先原样记录用户指令再解释执行；每轮会话按 HOLD→ACTIVE→DECISIONS→recent-work→MEMORY 启动，并给出文件冲突优先级与记忆评分入库规则。

justinhartbiz
justinhartbiz
热度 94.30
最近更新 2026-05-11 15:54:37
分类：知识依赖：读写文件应用场景：笔记整理 / 任务管理 / 工作流 / 模板

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 dory-memory
Metadata
name
dory-memory
description
File-based memory system for AI agents that forget between sessions. Implements the "Dory-Proof" pattern for continuity across context resets. Use when setting up agent memory, building workspace structure, implementing task tracking, or preventing context-loss errors. Triggers on "memory system", "remember between sessions", "Dory pattern", "agent continuity", or "workspace setup".

Dory-Proof Memory System

AI agents forget everything between sessions. This skill implements a file-based memory system that survives context resets.

Core Principle

Text > Brain. Write everything down. Files are memory. The agent only "remembers" what's on disk.

The Dory-Proof Pattern (Critical)

When the user gives a task:

  1. IMMEDIATELY write their EXACT WORDS to state/ACTIVE.md
  2. Then interpret what it means
  3. Then do the work
  4. Mark complete when done

Why: Paraphrasing introduces drift. Exact words preserve intent across context flushes.

Workspace Structure

workspace/
├── AGENTS.md        # Operating rules (system file, don't rename)
├── SOUL.md          # Identity + personality
├── USER.md          # About the human
├── MEMORY.md        # Curated long-term memory (<10KB)
├── LESSONS.md       # "Never again" safety rules
├── TOOLS.md         # Tool-specific notes
│
├── state/           # Active state (check every session)
│   ├── ACTIVE.md    # Current task (exact user words)
│   ├── HOLD.md      # Blocked items (check before acting!)
│   ├── STAGING.md   # Drafts awaiting approval
│   └── DECISIONS.md # Recent choices with timestamps
│
├── memory/          # Historical
│   ├── YYYY-MM-DD.md
│   ├── recent-work.md
│   └── archive/
│
└── ops/             # Operational
    └── WORKSPACE-INDEX.md

Boot Sequence (Every Session)

  1. Read state/HOLD.md — what's BLOCKED
  2. Read state/ACTIVE.md — current task
  3. Read state/DECISIONS.md — recent choices
  4. Read memory/recent-work.md — last 48 hours
  5. Read MEMORY.md — long-term (main session only)

Output status line after boot:

📋 Boot: ACTIVE=[task] | HOLD=[n] items | STAGING=[n] drafts

State File Formats

state/ACTIVE.md

## Current Instruction
**User said:** "[exact quote]"
**Interpretation:** [what you think it means]
**Status:**
- [ ] Step 1
- [ ] Step 2

state/HOLD.md

[YYYY-MM-DD HH:MM | session] Item — reason blocked

ALL agents must check before acting on anything that looks ready.

state/DECISIONS.md

[YYYY-MM-DD HH:MM | session] Decision made

Conflict Resolution

When files conflict, priority (highest first):

  1. state/HOLD.md — blocks override all
  2. state/ACTIVE.md — current instruction
  3. state/DECISIONS.md — recent choices
  4. AGENTS.md — general rules

Memory Scoring (Before Saving to MEMORY.md)

Score on 4 axes (0–3 each):

Axis 0 1 2 3
Longevity Gone tomorrow Weeks Months Years+
Reuse One-off Occasional Frequent Every session
Impact Trivial Nice to know Changes outputs Changes decisions
Uniqueness Obvious Slightly helpful Hard to rederive Impossible without

Save if: Total ≥ 8, OR any axis = 3 AND total ≥ 6.

Quick Setup

Copy template files from assets/templates/ to your workspace:

cp -r skills/dory-memory/assets/templates/* ~/.openclaw/workspace/

Then customize SOUL.md and USER.md for your agent.

References

  • references/IMPLEMENTATION-GUIDE.md — Full setup walkthrough
  • references/ANTI-PATTERNS.md — Common mistakes to avoid