multi-agent-brand-studio（多 Agent 品牌工作室）

name: multi-agent-brand-studio description: Use when setting up Multi-Agent Brand Studio on OpenClaw for multi-brand social media operations, approval-gated publishing, brand-isolated workspaces, or multi-agent content workflows. metadata: { "openclaw": { "emoji": "📱", "requires": { "bins": ["node"] } } }

Multi-Agent Brand Studio

Overview

This skill sets up Multi-Agent Brand Studio, a complete AI-powered social media operations team on OpenClaw. It creates:

• 5 specialized agents in a star topology (Leader + 4 specialists) + on-demand Reviewer
• Persistent A2A sessions for context-preserving multi-agent workflows
• 3-layer memory system (MEMORY.md + daily notes + shared knowledge base)
• Shared knowledge base with brand profiles, operations guides, and domain knowledge
• Approval workflow ensuring nothing publishes without owner approval
• Brand isolation with per-brand channels, content guidelines, and asset directories
• Cron automation for daily memory consolidation and weekly KB review

Optional Dependencies

• Image generation tool for Creator agent: The Creator agent requires an image generation tool installed in its workspace-creator/skills/ directory to produce images. Recommended: nano-banana-pro (Gemini-based, free tier). Without it, Creator produces text visual briefs only and cannot generate images.

Prerequisites

Before installing, ensure:

1. OpenClaw v2026.2.26+ is installed and openclaw onboard has been completed
2. At least one auth profile exists (e.g., Anthropic API key)
3. The ~/.openclaw/ directory exists

Quick Start

1. Install the skill (if not already in workspace/skills/)
2. Trigger setup: "Set up Multi-Agent Brand Studio"
3. Follow the interactive onboarding (6 steps, ~10 minutes)
4. Start creating content!

Onboarding Flow

When first triggered, this skill runs an interactive setup process.

Step 1: Prerequisites Check

Verify the environment is ready:

• [ ] OpenClaw installed and openclaw onboard completed
• [ ] ~/.openclaw/ directory exists
• [ ] At least one auth profile configured

If any prerequisite is missing, guide the user to resolve it before continuing.

Step 2: Team Setup

All 5 agents are installed automatically. Do not ask the user to choose a team size.

The full team:

On-demand:

Model — All agents inherit the model configured during openclaw onboard (at agents.defaults.model). No per-agent model setup is needed.

Advanced note: If you later want to run a leaner team, re-run scaffold.sh --agents leader,creator,engineer to scaffold a subset.

Step 3: Run Scaffold

Execute the setup scripts to create all directories and files first:

# 1. Create directories, copy templates, set up symlinks
bash scripts/scaffold.sh \
  --skill-dir "$(pwd)"

# 2. Merge agent configuration into openclaw.json
node scripts/patch-config.js \
  --config ~/.openclaw/openclaw.json

The scaffold creates:

• Agent workspace directories with SOUL.md, AGENTS.md, MEMORY.md
• Shared knowledge base with all template files
• Symlinks from each workspace to shared/
• Sub-skills (instance-setup, brand-manager) in Leader's skills/
• Cron job definitions

The config patcher merges into openclaw.json:

• Agent definitions with model assignments and tool restrictions
• A2A session configuration
• QMD memory paths (only if QMD is installed — otherwise skipped with a suggestion)
• Internal hooks

Step 4: Telegram Setup

This step uses a guided flow — do not ask the user for raw chat IDs or thread IDs.

Phase A: Confirm Bot Token

1. Check openclaw.json for channels.telegram.botToken
2. If present → skip to Phase B
3. If missing → guide the user:

- "Open Telegram, search for @BotFather" - "Send /newbot and follow the prompts to create a bot" - "Copy the bot token and paste it here" - Write the token into openclaw.json at channels.telegram.botToken

Phase B: Choose Channel Mode

Present the options in this order (Group+Topics first):

1. Group+Topics (recommended) — Best for most setups

- Brands are topic threads inside a Telegram supergroup - Works for both solo operators and multi-person teams - Requires a supergroup with Topics enabled

1. DM+Topics — Private alternative, no group needed

- Each brand gets its own topic thread inside the bot's DM - Requires enabling Thread Mode on the bot (guided below)

1. DM-simple — Minimal, no brand isolation

- Single DM conversation with the bot - Context-based brand routing (no topics)

1. Group-simple — Group without brand isolation

- Single group conversation - Context-based brand routing (no topics)

Phase C: Mode-Specific Setup

If DM+Topics:

1. Guide the user to enable Thread Mode on their bot:

- "Open Telegram, find @BotFather" - "Tap the Open button (bottom-left) to open the BotFather MiniApp" - "Select your bot in the MiniApp" - "Go to Bot Settings" - "Find Thread Mode and enable it" - "Come back and tell me when it's done"

1. Once confirmed, use the bot token to get the user's chat ID:

- "Send any message to your bot in Telegram" - Agent reads the incoming message context to extract the user's chat ID from {{From}} - Agent writes the chat ID into the channel config

1. Create the Operations topic automatically:

   node scripts/telegram-topics.js \
     --config ~/.openclaw/openclaw.json \
     --chat <USER_CHAT_ID> \
     --name "Operations"

1. Write the resulting thread ID into shared/operations/channel-map.md
2. Update cron/jobs.json — replace {{OPERATIONS_CHANNEL}} with the actual Operations channel address (format: chatId:threadId, e.g., 123456789:7)

If Group+Topics:

1. Check if the user already has a supergroup:

- If not: guide them to create one (Create Group → toggle "Topics" on)

1. Guide the user to add the bot to the group:

- "Add your bot to the supergroup" - "Make the bot an admin with the Manage Topics permission" - "Send a message in the group"

1. Agent reads the incoming message context to extract:

- Group chat ID from {{To}} - Agent writes the chat ID into the channel config

1. Create the Operations topic automatically:

   node scripts/telegram-topics.js \
     --config ~/.openclaw/openclaw.json \
     --chat <GROUP_CHAT_ID> \
     --name "Operations"

1. Write the resulting thread ID into shared/operations/channel-map.md
2. Update cron/jobs.json — replace {{OPERATIONS_CHANNEL}} with the actual Operations channel address (format: chatId:threadId, e.g., -100XXXXXXXXXX:7)

If DM-simple:

1. "Send any message to your bot in Telegram"
2. Agent reads the chat ID from the incoming message context
3. Write chat ID into channel config — done

If Group-simple:

1. Guide: "Add the bot to your group and send a message"
2. Agent reads the group chat ID from the incoming message context
3. Write chat ID into channel config — done

Step 5: Instance Setup + First Brand

After scaffolding and Telegram configuration, run the sub-skills:

1. Instance Setup (instance-setup skill)

- Owner name and timezone - Communication language (owner-facing) - Default content language - Bot identity (name, emoji, personality) - Updates: shared/INSTANCE.md, workspace/IDENTITY.md

1. First Brand (brand-manager add)

- Brand ID, display name, domain - Target market and content language - Topic creation (for Topics modes): - Agent calls scripts/telegram-topics.js to create a topic named after the brand - The script returns the thread ID - Agent writes the thread ID into shared/operations/channel-map.md and the brand config - For simple modes: no topic needed, skip thread ID - Creates: brand profile, content guidelines, domain knowledge file, asset directories

Step 6: Verification + Gateway Restart

1. Restart gateway:

   openclaw gateway restart

1. Run diagnostics:

   openclaw doctor

This validates: agent config, DM allowlist inheritance, session health, model availability, and workspace integrity.

1. Additional checks:

- [ ] Leader responds to messages - [ ] sessions_send to at least one agent succeeds

Optional: Enable QMD semantic memory

If patch-config.js reported "qmd binary not found" during Step 3, agents will use file-based memory (which works fine). To enable enhanced semantic search:

• Say "Set up QMD" to run the qmd-setup sub-skill, which guides you through installation and configuration.

Suggested first tasks after setup:

1. Fill in your brand profile: shared/brands/{brand_id}/profile.md
2. Test content creation: "Write a Facebook post for {brand}"
3. Add more brands: "Add a new brand"
4. Set up posting schedule: fill in shared/operations/posting-schedule.md

Post-Installation

Async Dispatch Model (v2.0.0+)

Leader uses fully async dispatch (sessions_send with timeoutSeconds: 0) for all agent communication. This means:

• Leader is never blocked waiting for an agent — always available to the owner.
• Agents callback to Leader via sessions_send when done (event-driven, not polling). Leader processes callbacks per the "Agent Callback Protocol" flow in AGENTS.md.
• Each task is tracked in a separate file: tasks/T-{YYYYMMDD}-{HHMM}.md. Completed tasks are archived to tasks/archive/.
• Stale task detection is handled by a cron job (stale-task-check, every 10 minutes) that scans tasks/ for steps stuck in [⏳] state. Runs as Leader.
• HEARTBEAT.md ships empty by default — periodic checks are handled by cron jobs instead of heartbeat polls.

Secrets Management (Optional)

For centralized API key management instead of scattered env vars:

openclaw secrets audit      # Check for plaintext secrets in config
openclaw secrets configure  # Set up secret entries
openclaw secrets apply      # Activate secrets
openclaw secrets reload     # Hot-reload without gateway restart

Adding More Brands

Use the brand-manager sub-skill:

• "Add a new brand" — interactive brand creation (auto-creates topic for Topics modes)
• "List brands" — show all active brands
• "Archive {brand}" — deactivate a brand

Customizing Agents

Each agent's behavior is defined in two files:

• SOUL.md — Persona, philosophy, boundaries, safety rules
• AGENTS.md — Operating procedures, data handling, brand scope, tools

Modify these files to tune agent behavior for your specific needs.

Memory System

The 3-layer memory system works automatically:

• MEMORY.md — Long-term curated memory (auto-updated by cron)
• memory/YYYY-MM-DD.md — Daily activity logs
• shared/ — Permanent knowledge base (grows over time)

Optional enhancement: Install QMD for semantic search across the knowledge base. Use the qmd-setup sub-skill or install manually (bun install -g @tobilu/qmd).

See references/memory-system.md for detailed documentation.

Communication Signals

Agents use standardized signals to communicate status. See references/signals-protocol.md for the complete signal dictionary.

Reference Documentation

Directory Structure

After installation, the following structure is created:

~/.openclaw/
├── openclaw.json                    # Updated with agent configs
├── workspace/                       # Leader
│   ├── SOUL.md, AGENTS.md, HEARTBEAT.md, IDENTITY.md
│   ├── memory/, skills/, assets/
│   └── shared/                      # Real directory (shared KB lives here)
│       ├── INSTANCE.md              # Instance configuration
│       ├── brand-registry.md        # Brand registry
│       ├── system-guide.md, brand-guide.md, compliance-guide.md
│       ├── team-roster.md
│       ├── brands/{id}/profile.md   # Per-brand profiles
│       ├── domain/{id}-industry.md  # Industry knowledge
│       ├── operations/              # Ops guides
│       └── errors/solutions.md      # Error KB
├── workspace-creator/               # Creator
│   ├── SOUL.md, AGENTS.md, MEMORY.md
│   ├── memory/, skills/
│   └── shared -> ../workspace/shared/
├── workspace-worker/                # Worker
│   └── (same structure)
├── workspace-researcher/            # Researcher
│   └── (same structure)
├── workspace-engineer/              # Engineer
│   └── (same structure)
├── workspace-reviewer/              # Reviewer (minimal, read-only)
│   ├── SOUL.md, AGENTS.md
│   └── shared -> ../workspace/shared/
└── cron/jobs.json                   # Scheduled tasks

Scripts

Sub-Skills