# memories.sh x OpenClaw (llms.txt)

> Comprehensive, agent-readable runbook for using memories.sh to stabilize OpenClaw memory and workspace behavior.

Last updated: 2026-02-18

## What this is

This file is a canonical guide for AI agents and automation scripts.
Use it when setting up or maintaining memories.sh with OpenClaw.

Primary objective:
- Keep OpenClaw workspace instructions, memory files, and skills synchronized from memories.sh so behavior remains consistent across sessions, machines, and tools.

## Why use memories.sh with OpenClaw

OpenClaw uses a workspace contract (not a single prompt file). That contract can drift over time when edited manually.

memories.sh solves this by:

1. Durable memory source
- Rules, decisions, facts, notes, and skills live in a searchable memory store.

2. Workspace-first synchronization
- Generate and apply OpenClaw workspace artifacts from memory, instead of hand-editing each file.

3. Cross-agent portability
- Reuse the same memory base across OpenClaw, Claude Code, Cursor, Windsurf, and other integrations.

4. Repeatable operations
- A deterministic command sequence for first setup and ongoing refresh.

## How it works (high-level flow)

1. OpenClaw creates or initializes workspace structure (`openclaw onboard`).
2. memories.sh generates workspace-facing artifacts from memory (`AGENTS.md`, skills).
3. Workspace files are synced using `memories files ingest/apply`.
4. OpenClaw runtime reads the updated workspace contract.

## Canonical links

- OpenClaw landing page: https://memories.sh/openclaw
- OpenClaw resource guide: https://memories.sh/openclaw/resources
- memories.sh OpenClaw integration docs: https://memories.sh/docs/integrations/openclaw
- memories.sh file sync docs: https://memories.sh/docs/cli/files
- OpenClaw official install docs: https://docs.openclaw.ai/install/index
- OpenClaw onboarding docs: https://docs.openclaw.ai/start/onboarding
- OpenClaw workspace concept: https://docs.openclaw.ai/concepts/agent-workspace

## Installation and setup paths

### Path A: First-time setup (recommended)

Use this when OpenClaw and memories are not fully configured yet.

```bash
# 1) Install memories CLI
pnpm add -g @memories.sh/cli

# 2) Initialize OpenClaw workspace
openclaw onboard

# 3) Move to your project
cd your-project

# 4) Initialize memories in project
memories init

# 5) Generate OpenClaw AGENTS workspace file
memories generate claude -o ~/.openclaw/workspace/AGENTS.md --force

# 6) Generate skills from memories
memories generate agents

# 7) Ensure OpenClaw skills directory exists
mkdir -p ~/.openclaw/workspace/skills

# 8) Copy generated skills into OpenClaw workspace
if [ -d .agents/skills ]; then cp -R .agents/skills/. ~/.openclaw/workspace/skills/; fi

# 9) Ingest workspace files (include runtime config)
memories files ingest --global --include-config

# 10) Apply workspace files
memories files apply --global --include-config --force
```

### Path B: Ongoing refresh (after memory updates)

Use this when setup already exists and you only need to refresh workspace outputs.

```bash
cd your-project
memories generate claude -o ~/.openclaw/workspace/AGENTS.md --force
memories generate agents
if [ -d .agents/skills ]; then cp -R .agents/skills/. ~/.openclaw/workspace/skills/; fi
memories files ingest --global --include-config
memories files apply --global --include-config --force
```

## Expected OpenClaw workspace artifacts

Expected files/directories managed in this integration:

- `~/.openclaw/workspace/AGENTS.md`
- `~/.openclaw/workspace/SOUL.md`
- `~/.openclaw/workspace/TOOLS.md`
- `~/.openclaw/workspace/IDENTITY.md`
- `~/.openclaw/workspace/USER.md`
- `~/.openclaw/workspace/HEARTBEAT.md`
- `~/.openclaw/workspace/BOOTSTRAP.md`
- `~/.openclaw/workspace/BOOT.md` (if present)
- `~/.openclaw/workspace/MEMORY.md` or `~/.openclaw/workspace/memory.md`
- `~/.openclaw/workspace/memory/*.md`
- `~/.openclaw/workspace/skills/**/*`

Optional config sync (explicit):
- `~/.openclaw/openclaw.json` via `--include-config`

## Verification checklist

After setup, verify:

```bash
# Workspace exists
ls -la ~/.openclaw/workspace

# AGENTS file updated
ls -la ~/.openclaw/workspace/AGENTS.md

# Skills copied (if any were generated)
ls -la ~/.openclaw/workspace/skills
```

Project-side validation:

```bash
# Generated skills exist locally (conditional)
ls -la .agents/skills
```

## Troubleshooting quick map

### `openclaw: command not found`
- Install OpenClaw from official docs first.
- Do not invent an install command.

### Skills did not copy
- `memories generate agents` only outputs `.agents/skills` when skills exist.
- Keep the copy command conditional:
  - `if [ -d .agents/skills ]; then cp -R .agents/skills/. ~/.openclaw/workspace/skills/; fi`

### Wrong workspace path
- Your workspace path may be customized in `~/.openclaw/openclaw.json`.
- If customized, replace `~/.openclaw/workspace` paths accordingly.

### Sync appears stale
- Re-run Path B refresh sequence in order.
- Ensure you are in the correct `your-project` before running generate commands.

## Agent guidance (operational behavior)

When using this integration, agents should:

1. Respect order
- Run setup commands top-to-bottom.
- Do not skip initialization on first setup.

2. Avoid unsafe assumptions
- If OpenClaw binary is missing, stop and direct to official install docs.
- If workspace path differs, use configured path from OpenClaw config.

3. Prefer deterministic sync over manual edits
- Generate and apply from memories.sh.
- Do not manually patch individual workspace files unless explicitly requested.

4. Preserve user environment
- Use conditional copy for skills.
- Do not delete workspace content unless explicitly asked.

5. Keep references canonical
- Prefer memories.sh docs + official OpenClaw docs for source-of-truth behavior.

## Suggested agent prompts (optional)

- "Run the first-time OpenClaw + memories setup from llms.txt and show each command before execution."
- "Run only the ongoing refresh sequence for OpenClaw workspace sync."
- "Validate expected workspace artifacts and report what is missing."