Skills skill

Name: skill
Author: openclaw

AVM Memory Skill

install

source · Clone the upstream repo

git clone https://github.com/openclaw/skills

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/openclaw/skills "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/bkmashiro/vfs/skill" ~/.claude/skills/openclaw-skills-skill-d8f47b && rm -rf "$T"

OpenClaw · Install into ~/.openclaw/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/openclaw/skills "$T" && mkdir -p ~/.openclaw/skills && cp -r "$T/skills/bkmashiro/vfs/skill" ~/.openclaw/skills/openclaw-skills-skill-d8f47b && rm -rf "$T"

manifest: skills/bkmashiro/vfs/skill/SKILL.md

AVM Memory Skill

AI Virtual Memory for agents. Token-aware recall, knowledge graphs, multi-agent support.

Quick Start

1. Configure AVM Daemon

# ~/.config/avm/mounts.yaml
mounts:
  - mountpoint: ~/.openclaw/workspace-{agent}/avm
    agent_id: {agent}

# Start daemon
avm-daemon start --daemon

# Verify
ls ~/.openclaw/workspace-{agent}/avm/

2. Basic Usage

# Store a memory
echo "RSI > 70 indicates overbought" > avm/notes/rsi_rule.md

# Recall with query
cat "avm/:recall?q=RSI trading"

# See what's in memory
cat avm/:topics

# Browse structure
cat avm/:list

3. Virtual Paths

Path	Description
`avm/:recall?q=...`	Semantic search
`avm/:topics`	Topic overview
`avm/:list`	List all memories
`avm/:handlers`	Available handlers
`avm/:handlers/{name}`	Specific handler docs
`avm/notes/*.md`	Read/write memories

For Sub-Agents

When spawning a sub-agent that needs memory:

Step 1: Add to mounts.yaml

# ~/.config/avm/mounts.yaml
mounts:
  - mountpoint: ~/.openclaw/workspace-{subagent}/avm
    agent_id: {subagent}

# Reload daemon
avm-daemon reload

Step 2: Add to AGENTS.md

## Memory

This agent has AVM memory mounted at `avm/`.

### Recall
\`\`\`bash
cat "avm/:recall?q=your query"
\`\`\`

### Remember
\`\`\`bash
echo "content" > avm/notes/title.md
\`\`\`

### Topics
\`\`\`bash
cat avm/:topics
\`\`\`

Step 3: Verify

# Test recall
cat "avm/:recall?q=test"

Commands

# Daemon management
avm-daemon start --daemon  # Start in background
avm-daemon status          # Show mounts
avm-daemon reload          # Reload config
avm-daemon stop            # Stop all

# CLI operations
avm recall "query" --agent {id}
avm remember "content" --title "name" --agent {id}
avm topics --agent {id}

# Import existing memories
avm import /path/to/*.md --agent {id}

Multi-Agent Setup

Shared Namespace

Multiple agents can share memories:

# Agent A writes to shared
echo "market signal" > avm/shared/market/signal.md

# Agent B reads shared
cat "avm/:recall?q=market signal"  # Finds it

Private Memories

Each agent's writes go to their private namespace by default:

avm/
├── private/{agent}/*.md   # Only this agent
└── shared/{namespace}/*.md # All agents

Tips

Write in English - FTS has no Chinese tokenizer; English searches work better
Use recall, not search -
```
cat avm/:recall?q=...
```
ranks by relevance + importance
Use delta for collaboration -
```
cat shared/file.md:delta
```
shows only changes since last read
Set importance - High importance memories surface first
Use tags - Add tags via file content:
```
tags: [trading, risk]
```
Link memories - Use
```
avm link
```
to build knowledge graph
Check topics first -
```
cat avm/:topics
```
to see what's available

Troubleshooting

Mount fails

# Check daemon status
avm-daemon status

# Restart daemon
avm-daemon stop && avm-daemon start --daemon

# Force unmount (macOS)
umount -f /path/to/mount

# Force unmount (Linux)
fusermount -uz /path/to/mount

Permission denied

# Check ownership
ls -la ~/.local/share/avm/

# Fix if needed
chown -R $USER ~/.local/share/avm/

No results

# Check if memories exist
avm list --agent {id}

# Check topics
cat avm/:topics