Memory Transfer Enhanced

Transfer memories between OpenClaw agents with optional topic filtering, supporting sharing mode with role transformation or cloning mode for verbatim copies.

Security Scan

The skill's behavior mostly matches its memory-transfer description, but there are inconsistencies (missing declared runtime, metadata mismatches, and file-system access) that deserve caution before installing.

Purpose & Capability

The code and SKILL.md implement memory listing, topic filtering, share vs clone modes, and file copying between agent workspaces under /home/node/.openclaw — this aligns with the stated purpose. Minor inconsistencies: the registry lists the skill as 'memory-transfer-enhanced' v2.1.0 while _meta.json inside the package claims v1.0.0 and a strange publishedAt timestamp. Also the SKILL.md and code invoke 'node' but the skill declares no required binaries.

Instruction Scope

Runtime instructions and the included script explicitly read agent workspace directories, read specific memory files, optionally filter/transform content, and write into a target agent memory directory. There are no network calls, external endpoints, or instructions to read unrelated system areas; prompts and a dry-run/confirm step are included.

Install Mechanism

There is no install spec (instruction-only), but a shipped executable JS file (memory-transfer.js) expects Node to be present and is intended to be run directly. The skill metadata did not declare Node or any required binary, which is an omission and a mismatch between declared requirements and actual runtime needs.

Credentials

The skill requests no environment variables or external credentials, and it does not call out network endpoints. However, it reads and writes files under /home/node/.openclaw for arbitrary agent IDs and will therefore have access to all agent memories on the host; this access is necessary for the stated purpose but is broad — verify you trust the skill and the environment where it will run.

Persistence & Privilege

The skill is not marked always:true and does not claim to modify other skills or system-wide configuration. It writes copied/translated memory files into target workspaces (expected behavior for a transfer tool).

What to consider before installing

This package largely does what its description says (reads agent memory files, filters/transforms them, and writes them into a target agent workspace), but take these precautions before installing or running it: - Runtime requirement: The tool is a Node script but the skill metadata does not declare Node as a required binary — ensure Node is present and you understand that the script will run on your host. - Source trust: The package has no homepage and the _meta.json version/timestamp mismatch is odd; only run code from authors you trust. Inspect the full memory-transfer.js source yourself (you should review the truncated tail to confirm there are no hidden network calls or obfuscated logic). - Dry-run & backup: Use the --dry-run and interactive confirm flow first. Back up any target agent workspace before performing a non-dry run to avoid accidental overwrites. - Least privilege / sandboxing: Run the tool in a restricted environment or container if possible, since it reads/writes all agent workspaces under /home/node/.openclaw. - Review transformations: The 'share' mode attempts automatic pronoun/name transformations and regex-based user-data filtering; these are heuristic and may remove or alter content unexpectedly. Verify results in the preview before committing. If you need higher assurance, ask the publisher for provenance (homepage, source repo, signed release) or request that the skill declare required binaries (node) and provide a reproducible install path.

Memory Transfer Skill

Transfer memory files and context between OpenClaw agents with advanced filtering and transformation options.

Core Features

1. Topic-Specific Transfer - Transfer only memories related to a specific topic/keyword
2. Memory Sharing - Share memories with role/perspective transformation (I → you, my → your)
3. Memory Cloning - Clone memories verbatim without any transformation

Mode Differences

Memory Sharing (share mode)

Best for knowledge/context transfer between agents

This mode:

1. Filters out user information - Removes personal data about the user:

   • User names: "我叫小明", "我的名字是..."
   • User preferences: "我喜欢...", "我讨厌..."
   • User personal info: email, phone, address, birthday
   • About user: "关于我..."

2. Transforms to target agent's identity - Converts references to match the target agent:

   • "I am Agent Main" → "I am Agent Skill Master" (adopts target identity)
   • "Agent Main's workspace" → "Agent Skill Master's workspace"
   • "I work as a helper" → "I work as Agent Skill Master"
   • First-person pronouns (I, my, me) remain as first-person

Use when: Sharing project knowledge, workflows, or task context. The target agent adopts the knowledge as their own experience.

Memory Cloning (clone mode)

Copies memory exactly as-is without filtering or transformation:

• All content preserved verbatim
• User information remains
• Original context maintained

Use when: Full backup/migration or preserving original author's voice.

Commands

List available agents

ls /home/node/.openclaw/

Transfer all memory from source to target

node memory-transfer.js transfer <source-agent-id> <target-agent-id>

Transfer with mode selection (interactive)

node memory-transfer.js transfer <source> <target> --mode interactive

This will prompt for:

1. Topic filter (optional - press Enter to transfer all)
2. Mode selection: share or clone
3. Confirmation before proceeding

Transfer specific topic memory

node memory-transfer.js transfer <source> <target> --topic "claude"

Force memory sharing (with transformation)

node memory-transfer.js transfer <source> <target> --mode share

Force memory cloning (verbatim)

node memory-transfer.js transfer <source> <target> --mode clone

Preview transfer (dry run)

node memory-transfer.js transfer <source> <target> --dry-run

List agent memories

node memory-transfer.js list <agent-id>

Search memories by topic

node memory-transfer.js search <agent-id> <topic>

Interactive Workflow

When running without explicit flags, the skill will:

1. Prompt for source agent - Select which agent's memory to transfer
2. Prompt for target agent - Select destination agent
3. Prompt for topic (optional) - Enter keyword to filter, or press Enter for all
4. Prompt for mode - Choose:
   • share - Filter user info + transform pronouns (recommended for knowledge sharing)
   • clone - Keep original verbatim (for backup)
5. Show preview - Display what will be transferred
6. Confirm - Require explicit confirmation before executing
7. Execute or cancel - Proceed or abort based on user decision

Examples

Transfer all memories as sharing (with transformation)

node memory-transfer.js transfer main coder --mode share

Transfer specific date memory as clone

node memory-transfer.js transfer main coder 2026-03-01.md --mode clone

Transfer topic-filtered memory with preview

node memory-transfer.js transfer main coder --topic "preferences" --dry-run

Interactive mode (recommended for first-time use)

node memory-transfer.js transfer main coder --mode interactive

Target Agent Adaptation Rules

When using share mode, the memory is transformed to match the target agent's identity:

Agent Identity

Source	Target
I	Agent Main (source agent name)
me	Agent Main
my	Agent Main's
I bought	Agent Main bought
I think	Agent Main thinks

Role Statements

Source Pattern	Target
I work as helper	Agent Skill Master works as helper
I serve as	Agent Skill Master serves as

Result example: "I bought a phone" → "Agent Main bought a phone"

This way the target agent knows this is the source agent's experience, not their own.

Source	Target
I think	you think
I believe	you believe
I know	you know
I understand	you understand
I remember	you remember
I prefer	you prefer
I like	you like
I want	you want
I need	you need

Important: The transformation converts both pronouns AND role descriptions so the target agent doesn't inherit confused identity. For example:

• "I am Agent Skill Master" → "you are Agent Skill Master"
• "My role is to create skills" → "your role is to create skills"
• "I was created to help you" → "you were created to help me"

Agent Workspaces

OpenClaw agent workspaces are typically at:

• /home/node/.openclaw/workspace-<agent-id>/
• Main agent: /home/node/.openclaw/workspace-main/

Memory files:

• MEMORY.md - Long-term memory
• memory/YYYY-MM-DD.md - Daily memories
• memory/*.md - Topic-specific memories

Safety Features

1. Dry-run by default - Preview before executing
2. Explicit confirmation - Never auto-execute without approval
3. Backup option - Can create backup before overwriting
4. Mode clarification - Always ask to confirm share vs clone mode