---

name: honcho-setup description: > Install the @honcho-ai/openclaw-honcho plugin and migrate legacy file-based memory to Honcho. UPLOADS WORKSPACE CONTENT TO EXTERNAL API: Sends USER.md, MEMORY.md, IDENTITY.md, memory/, canvas/, SOUL.md, AGENTS.md, BOOTSTRAP.md, TOOLS.md, and HEARTBEAT.md contents to api.honcho.dev (managed, default) or your self-hosted HONCHO_BASE_URL. Requires user confirmation before uploading. Archives originals locally with user confirmation. Updates workspace docs to reference Honcho tools. Works with managed Honcho (requires API key) or self-hosted local instances (no key needed). metadata: openclaw: emoji: "🧠" required_env: [] # Nothing is strictly required - self-hosted mode works without API key optional_env: - name: HONCHO_API_KEY description: "REQUIRED for managed Honcho (https://app.honcho.dev). NOT required for self-hosted instances. This skill reads this value from environment variables or ~/.openclaw/.env file." - name: HONCHO_BASE_URL description: "Base URL for a self-hosted Honcho instance (e.g. http://localhost:8000). Defaults to https://api.honcho.dev (managed)." - name: HONCHO_WORKSPACE_ID description: "Honcho workspace ID. Defaults to 'openclaw'." - name: WORKSPACE_ROOT description: "Path to the OpenClaw workspace root. Auto-detected from ~/.openclaw/openclaw.json if not set." required_binaries: - node - npm optional_binaries: - git - docker - docker-compose writes_to_disk: true archive_directory: "{workspace_root}/archive/" reads_sensitive_files: - "~/.openclaw/.env - reads ONLY the HONCHO_API_KEY value, no other environment variables are accessed" - "~/.openclaw/openclaw.json - reads workspace path configuration only" network_access: - "api.honcho.dev (managed mode, default)" - "User-configured HONCHO_BASE_URL (self-hosted mode)" data_handling: uploads_to_external: true requires_user_confirmation: true external_destinations: - "api.honcho.dev (managed Honcho, default)" - "User-configured HONCHO_BASE_URL (self-hosted mode)" uploaded_content: - "USER.md, MEMORY.md, IDENTITY.md (user profile/memory files)" - "All files under memory/ directory (structured memory)" - "All files under canvas/ directory (working memory)" - "SOUL.md, AGENTS.md, BOOTSTRAP.md, TOOLS.md, HEARTBEAT.md (agent configuration)" data_destination_purpose: "Migrates file-based memory system to Honcho API for AI agent memory/personalization" homepage: "https://honcho.dev" source: "https://github.com/plastic-labs/honcho"

Honcho Setup

Install the Honcho plugin and migrate legacy workspace memory files to Honcho.

⚠️ DATA UPLOAD WARNING: This skill uploads the contents of your workspace memory files (USER.md, MEMORY.md, IDENTITY.md, memory/, canvas/, SOUL.md, AGENTS.md, BOOTSTRAP.md, TOOLS.md, HEARTBEAT.md) to an external API. By default, data is sent to api.honcho.dev (managed Honcho cloud service). For self-hosted instances, data is sent to your configured HONCHO_BASE_URL. You will be asked for explicit confirmation before any upload occurs, and you will see exactly which files will be uploaded and where they will be sent.

This skill modifies workspace files. It will ask for confirmation before archiving or deleting any files. If the Honcho upload fails or is skipped, no files are moved or removed. All files are backed up to {workspace_root}/archive/ before any deletion.

Sensitive file access: This skill reads ~/.openclaw/.env to check for HONCHO_API_KEY only (required for managed Honcho). No other environment variables are read from this file. It also reads ~/.openclaw/openclaw.json to determine workspace location.

Step 1: Install and Enable the Plugin

Install the Honcho plugin using the OpenClaw plugin system. Use this exact command — do not install @honcho-ai/sdk directly or use npm install in the workspace.

openclaw plugins install @honcho-ai/openclaw-honcho

Then enable it:

openclaw plugins enable openclaw-honcho

After enabling, verify the plugin loaded without errors. If the gateway logs show Cannot find module '@honcho-ai/sdk', the plugin's dependencies need to be installed manually:

cd ~/.openclaw/extensions/openclaw-honcho && npm install

Then restart the gateway. This is a known issue with the OpenClaw plugin installer not running dependency resolution for plugin packages.

If the plugin is already installed and enabled, skip to Step 2.

Step 2: Verify Honcho Connection

Honcho can run as a managed cloud service or as a self-hosted local instance. Determine which the user is using.

Option A: Managed Honcho (default)

Confirm that HONCHO_API_KEY is set. Check the environment variables first. If not found, read ONLY the HONCHO_API_KEY value from ~/.openclaw/.env if that file exists. Do not read or access any other environment variables from the .env file — only extract the HONCHO_API_KEY value needed for this migration.

If the key is not set in either location, stop and tell the user:

HONCHO_API_KEY is not set. Add it to your environment or ~/.openclaw/.env, then re-run this skill. You can get a key at https://app.honcho.dev

Option B: Self-hosted / local Honcho

Honcho is open source and can be run locally. If the user is running their own instance, they need to set HONCHO_BASE_URL to point to it (e.g., http://localhost:8000). The SDK environment should be set to "local".

A local instance can be started with docker-compose from the Honcho repo (requires git, docker, and docker-compose):

git clone https://github.com/plastic-labs/honcho
cd honcho
cp .env.template .env
cp docker-compose.yml.example docker-compose.yml
docker compose up

For local instances, HONCHO_API_KEY may not be required depending on the user's configuration. Verify connectivity before proceeding.

See https://github.com/plastic-labs/honcho?tab=readme-ov-file#local-development for full self-hosting instructions.

Do not proceed with migration until the connection is verified. No files will be read, uploaded, archived, or removed without a working Honcho connection.

Step 3: Detect Legacy Memory Files

Scan the workspace root for legacy memory files. The workspace root is determined by (in priority order):

1. The WORKSPACE_ROOT environment variable
2. The agent.workspace or agents.defaults.workspace field in ~/.openclaw/openclaw.json
3. ~/.openclaw/workspace

Files to detect

User/owner files (content describes the user):

• USER.md
• IDENTITY.md
• MEMORY.md

Agent/self files (content describes the agent):

• SOUL.md
• AGENTS.md
• TOOLS.md
• BOOTSTRAP.md
• HEARTBEAT.md

Directories:

• memory/ — recursively read all files
• canvas/ — recursively read all files

Files inside memory/ and canvas/ are treated as user/owner content.

Report what was found to the user before proceeding. IMPORTANT: You MUST ask for explicit confirmation before proceeding.

When asking for confirmation, provide this exact information to the user:

Found legacy memory files ready for migration:

• [List each file found with its size]

What will happen next if you confirm:

1. Upload: All file contents will be uploaded to [api.honcho.dev OR your self-hosted URL]
2. Archive: Files will be copied to {workspace_root}/archive/ for backup
3. Remove: Legacy-only files (USER.md, MEMORY.md, IDENTITY.md, HEARTBEAT.md, memory/, canvas/) will be removed after successful archive
4. Update: Workspace docs (SOUL.md, AGENTS.md, BOOTSTRAP.md) will be updated to use Honcho tools

Data destination: Your file contents will be sent to [show actual URL based on HONCHO_BASE_URL config]

Do you want to proceed with this migration?

Do not proceed to Step 4 without explicit user confirmation.

Step 4: Upload to Honcho

Upload each detected file to Honcho using the messages upload endpoint (or honcho_analyze if available):

• User/owner files → upload as messages in a session, using the owner peer (peer_id = owner peer id).
• Agent/self files → upload as messages in a session, using the openclaw peer (peer_id = openclaw peer id).

Ensure the workspace and both peers exist (e.g. via SDK or API) before uploading. Get or create a session for the uploads. Report how many files were uploaded for each category (user vs. agent).

If any upload fails, stop and warn the user. Do not proceed to archiving.

SDK setup example (messages upload with file)

Use the Honcho SDK to create messages from each file via the session upload API (the same operation as the REST .../messages/upload endpoint with file and peer_id). Set up the client and peers, get or create a session, add both peers to the session, then upload each detected file with the appropriate peer.

Note: The workspaceId and session name below are defaults. Customize them via the HONCHO_WORKSPACE_ID env var or pass your own session name if you manage multiple migrations.

import fs from "fs";
import path from "path";
import { Honcho } from "@honcho-ai/sdk";

const apiKey = process.env.HONCHO_API_KEY;
const workspaceRoot = process.env.WORKSPACE_ROOT || "~/.openclaw/workspace";

const honcho = new Honcho({
  apiKey,
  baseURL: process.env.HONCHO_BASE_URL || "https://api.honcho.dev",
  // Customize via HONCHO_WORKSPACE_ID or leave as default
  workspaceId: process.env.HONCHO_WORKSPACE_ID || "openclaw",
});

await honcho.setMetadata({});
const openclawPeer = await honcho.peer("openclaw", { metadata: {} });
const ownerPeer = await honcho.peer("owner", { metadata: {} });

// Session name can be customized for multiple migration runs
const session = await honcho.session("migration-upload", { metadata: {} });
await session.addPeers([ownerPeer, openclawPeer]);

// For each detected file: read file and call session.uploadFile(file, peer)
// User/owner files → ownerPeer; agent/self files → openclawPeer
const filesToUpload = [
  { path: path.join(workspaceRoot, "USER.md"), peer: ownerPeer },
  { path: path.join(workspaceRoot, "SOUL.md"), peer: openclawPeer },
  // ... add every detected file and files under memory/ and canvas/
];

for (const { path: filePath, peer } of filesToUpload) {
  const stat = await fs.promises.stat(filePath).catch(() => null);
  if (!stat?.isFile()) continue;
  const filename = path.basename(filePath);
  const content = await fs.promises.readFile(filePath);
  const content_type = "text/markdown"; // or "text/plain", "application/pdf", "application/json"
  const messages = await session.uploadFile(
    { filename, content, content_type },
    peer,
    {}
  );
  console.log(`Uploaded ${filePath}: ${messages.length} messages`);
}

• Required: session.uploadFile(file, peer, options?) — second argument is the peer (object or id). Use the owner peer for user/owner files (USER.md, IDENTITY.md, MEMORY.md, and everything under memory/ and canvas/), and the openclaw peer for agent/self files (SOUL.md, AGENTS.md, TOOLS.md, BOOTSTRAP.md, HEARTBEAT.md).
• Session: Add both peers to the session with session.addPeers([ownerPeer, openclawPeer]) before uploading.
• File (Node): Pass { filename, content: Buffer | Uint8Array, content_type }. See Honcho file uploads for supported types (PDF, text, JSON). A runnable test script is in scripts/test-upload-file.mjs (requires HONCHO_API_KEY).

Step 5: Archive Legacy Files

CRITICAL: Ask the user for explicit confirmation before archiving. The default archive location is {workspace_root}/archive/. The user may choose a different directory.

Safety guarantee: No file will ever be deleted without a backup copy existing in the archive directory first.

For each detected file:

1. Create the archive directory if it does not exist.
2. Copy the file into the archive directory. Verify the copy succeeded before proceeding.
3. If a file with the same name already exists in archive, append a timestamp (e.g., USER.md-2026-02-10T22-55-12).
4. Only after successful copy verification, apply the removal rules below.

Then apply these rules:

Remove originals after archiving (legacy-only files, no longer needed once migrated to Honcho):

• USER.md
• MEMORY.md
• IDENTITY.md
• HEARTBEAT.md

Keep originals in place (these are active workspace docs updated in Step 6):

• AGENTS.md
• TOOLS.md
• SOUL.md
• BOOTSTRAP.md

Move directories into the archive (contents already uploaded to Honcho):

• memory/
• canvas/

No files are deleted without a backup existing in the archive directory first. Every removal is preceded by a confirmed copy.

If the upload in Step 4 failed or was skipped, do not archive or remove any files. Warn the user that all files are preserved to prevent data loss.

Step 6: Update Workspace Docs

The plugin ships template files in node_modules/@honcho-ai/openclaw-honcho/workspace_md/. Use these templates as the source of truth for Honcho-aware workspace docs.

For each of AGENTS.md, SOUL.md, and BOOTSTRAP.md:

• If the file exists in the workspace: update it — replace references to the old file-based memory system (USER.md, MEMORY.md, memory/ directory, manual file reads/writes for memory) with Honcho tool references.
• If the file does not exist: copy the template into the workspace.
• Preserve any custom content the user has added. Only update memory-related sections.

The Honcho tools are: honcho_profile, honcho_context, honcho_search, honcho_recall, honcho_analyze.

Step 7: Confirm

Summarize what happened:

• Which legacy files were found
• How many files were uploaded (user and agent counts)
• Which files were archived and where
• Which workspace docs were created or updated
• That Honcho is now the memory system — no more manual file management needed

Provide a link to the Honcho docs for reference: https://docs.honcho.dev

---

Security & Privacy Disclosure

This skill has been designed with transparency and safety as priorities. Below is a complete disclosure of what this skill does:

Data Upload

• What is uploaded: Contents of USER.md, MEMORY.md, IDENTITY.md, all files under memory/, all files under canvas/, SOUL.md, AGENTS.md, BOOTSTRAP.md, TOOLS.md, HEARTBEAT.md
• Where it goes: By default to api.honcho.dev (managed Honcho cloud service). For self-hosted instances, to your configured HONCHO_BASE_URL
• User control: Explicit confirmation required before any upload. You will see the exact list of files and the destination URL
• Purpose: Migrating file-based memory system to Honcho API for AI agent personalization and memory

Sensitive File Access

• ~/.openclaw/.env: This skill reads ONLY the HONCHO_API_KEY value from this file (if present). No other environment variables are read or accessed
• ~/.openclaw/openclaw.json: This skill reads workspace path configuration only (agent.workspace or agents.defaults.workspace fields)
• Workspace files: All legacy memory files listed above are read for upload

File Modifications

• Archives: Creates {workspace_root}/archive/ and copies all files there before any deletion
• Deletions: USER.md, MEMORY.md, IDENTITY.md, HEARTBEAT.md, memory/, canvas/ are deleted ONLY after successful archive copy verification
• Updates: SOUL.md, AGENTS.md, BOOTSTRAP.md are updated to reference Honcho tools (preserved in archive first)
• Safety: No file is ever deleted without a verified backup copy existing first

Credentials

• HONCHO_API_KEY: Required only for managed Honcho (api.honcho.dev). Not required for self-hosted instances
• No other credentials: This skill does not access, read, or transmit any other credentials or secrets

Network Access

• Managed mode: Connects to api.honcho.dev (Honcho cloud service)
• Self-hosted mode: Connects to your configured HONCHO_BASE_URL (e.g., http://localhost:8000)
• Protocol: All uploads use the Honcho SDK (@honcho-ai/sdk) via the messages upload endpoint

User Control

• Step 3: Explicit confirmation required before any file upload (shows file list and destination URL)
• Step 5: Explicit confirmation required before any file archiving/deletion
• Step 6: Workspace doc updates preserve custom content
• Failure handling: If upload fails, no files are archived or deleted

Data Retention

• Local backups: All original files are preserved in {workspace_root}/archive/ indefinitely
• Remote storage: Uploaded data is stored according to Honcho's data retention policy (see https://honcho.dev/privacy)
• Self-hosted control: If using self-hosted Honcho, you control all data retention

Open Source

• Honcho SDK: Open source at https://github.com/plastic-labs/honcho
• Plugin code: Available at ~/.openclaw/extensions/openclaw-honcho after installation
• This skill: You are reading the complete skill instructions - there is no hidden behavior