OpenClaw Multi Agent Org

Use this skill to create or maintain a company-style OpenClaw multi-agent organization.

It creates:

• A CEO strategic hub plus functional agents for BD, media, content, editors, community, and ops.
• One isolated workspace per agent.
• Rich role files: AGENT.md, README.md, IDENTITY.md, SOUL.md, AGENTS.md, USER.md, TOOLS.md, TASKS.md, HEARTBEAT.md, BOOTSTRAP.md.
• OpenClaw config entries in ~/.openclaw/openclaw.json.
• Optional Feishu/Lark channels.feishu.accounts entries and bindings so each Feishu accountId routes to the matching OpenClaw agentId.

Fast Path

Inspect first:

openclaw config validate
openclaw agents list --bindings

Create or refresh the org locally:

node <skill-dir>/scripts/create-multi-feishu-agents.mjs --dry-run --skip-ceo
node <skill-dir>/scripts/create-multi-feishu-agents.mjs --skip-ceo

--dry-run redacts secrets by default. Use --dry-run-raw only when local debugging requires raw credentials.

Regenerate role/persona files as well:

node <skill-dir>/scripts/create-multi-feishu-agents.mjs --role-files refresh --skip-ceo

Validate and restart:

openclaw config validate
openclaw agents list --bindings
openclaw gateway restart

Feishu One-Click Flow

Official Feishu one-click creation usually writes the newly created app into top-level channels.feishu.appId/appSecret. After the user scans and confirms, adopt that app into the intended account:

node <skill-dir>/scripts/create-multi-feishu-agents.mjs --adopt-top-level content-chief --restart

This will:

• Copy the top-level Feishu app credentials into channels.feishu.accounts.content-chief.
• Set the target account enabled=true.
• Restore the top-level/default Feishu app from the CEO account when available.
• Remove incompatible tools.alsoAllow values sometimes written by installer flows.
• Validate the config, and restart if --restart is passed.

For repeated app creation, use one account at a time:

npx --yes @larksuite/openclaw-lark-tools install --verbose --secret-input-mode plaintext --skip-version-check
node <skill-dir>/scripts/create-multi-feishu-agents.mjs --adopt-top-level media-lead --restart

Or use the wrapper command, which runs the official installer first and then adopts the returned top-level app into the target account:

node <skill-dir>/scripts/create-multi-feishu-agents.mjs --feishu-one-click media-lead --restart

Read references/feishu-one-click.md if you need the operator checklist or troubleshooting notes.

Credentials File Mode

Use this when credentials are already known:

node <skill-dir>/scripts/create-multi-feishu-agents.mjs --credentials /path/to/feishu-apps.json --skip-ceo --restart

Shape:

{
  "ai-bd": {
    "appId": "cli_xxx",
    "appSecret": "xxx",
    "enabled": true
  }
}

Example: references/feishu-apps.example.json.

Custom Roles

Use a roles file when the user wants a different org chart:

node <skill-dir>/scripts/create-multi-feishu-agents.mjs --roles /path/to/roles.json --dry-run
node <skill-dir>/scripts/create-multi-feishu-agents.mjs --roles /path/to/roles.json --role-files refresh

Example: references/roles.example.json.

Script Options

Primary script:

scripts/create-multi-feishu-agents.mjs

Common options:

• --dry-run: print the planned config without writing.
• --dry-run-raw: print the planned config without writing, including secrets.
• --interactive: prompt for App ID / App Secret.
• --credentials PATH: read app credentials from JSON.
• --roles PATH: read custom role definitions from JSON.
• --only ACCOUNT_ID: update only one role/account/binding.
• --feishu-one-click ACCOUNT_ID: run the official Feishu installer, then adopt the created top-level app into the named account.
• --adopt-top-level ACCOUNT_ID: move the current top-level Feishu app into one account after scan-to-create.
• --skip-ceo: keep CEO/default credentials unchanged.
• --role-files missing|refresh|none: create missing files, overwrite generated role files, or skip role files.
• --no-validate: skip openclaw config validate.
• --restart: restart OpenClaw gateway after successful write.

Default Roles

Default paths:

~/.openclaw/agents/company-org/<agentId>/
~/.openclaw/workspace/company-org/<role-workspace>/

Safety Rules

• Always back up ~/.openclaw/openclaw.json before writing; the script does this automatically.
• Do not print or paste App Secret values into chat responses.
• Do not enable placeholder accounts whose credentials still start with __FILL_.
• Keep detailed role/persona content in Markdown files, not arbitrary fields in agents.list; OpenClaw config schema is strict.
• Validate after every write.
• Preserve user-customized role files unless the user explicitly wants regeneration; default --role-files missing is conservative.

Publishing Notes

For publishing or packaging this skill, keep SKILL.md concise and put operator details in references/. Human-facing overview is in README.md; agent-facing execution rules remain here.