social-media-ops
Set up a complete multi-brand social media management team on OpenClaw. Scaffolds 5 specialized agents (Leader, Creator, Worker, Researcher, Engineer) + on-demand Reviewer in a star topology with persistent A2A sessions, 3-layer memory system, shared knowledge base, approval workflows, and brand isolation. Use when setting up a new social media operations team, adding the multi-agent framework to an existing OpenClaw instance, or when the user mentions social media management, multi-brand operations, or content team setup.
- License 许可
- MIT
- 最近更新
- 2026-06-07
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- social-media-ops
- metadata
- { "openclaw": { "emoji": "📱", "requires": { "bins": [ "node" ] } } }
- description
- Set up a complete multi-brand social media management team on OpenClaw. Scaffolds 5 specialized agents (Leader, Creator, Worker, Researcher, Engineer) + on-demand Reviewer in a star topology with persistent A2A sessions, 3-layer memory system, shared knowledge base, approval workflows, and brand isolation. Use when setting up a new social media operations team, adding the multi-agent framework to an existing OpenClaw instance, or when the user mentions social media management, multi-brand operations, or content team setup.
03 SKILL.md 正文
Social Media Ops
Overview
This skill sets up 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:
- OpenClaw v2026.2.26+ is installed and
openclaw onboardhas been completed - At least one auth profile exists (e.g., Anthropic API key)
- The
~/.openclaw/directory exists
Quick Start
1. Install the skill (if not already in workspace/skills/)
2. Trigger setup: "Set up my social media operations team"
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 onboardcompleted - [ ]
~/.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:
| Agent | Role |
|---|---|
| Leader | Orchestration, routing, quality gates |
| Creator | Content + visual (copywriting, image gen, platform formatting) |
| Worker | Execution for Leader (files, CLI, config, maintenance) |
| Researcher | Market research, competitor analysis |
| Engineer | Technical integrations, automation |
On-demand:
| Agent | Role |
|---|---|
| Reviewer | Independent quality review (spawned when needed) |
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,engineerto 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
- Check
openclaw.jsonforchannels.telegram.botToken - If present → skip to Phase B
- If missing → guide the user:
- "Open Telegram, search for @BotFather"
- "Send
/newbotand follow the prompts to create a bot" - "Copy the bot token and paste it here"
- Write the token into
openclaw.jsonatchannels.telegram.botToken
Phase B: Choose Channel Mode
Present the options in this order (Group+Topics first):
-
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
-
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)
-
DM-simple — Minimal, no brand isolation
- Single DM conversation with the bot
- Context-based brand routing (no topics)
-
Group-simple — Group without brand isolation
- Single group conversation
- Context-based brand routing (no topics)
Phase C: Mode-Specific Setup
If DM+Topics:
- 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"
- 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
- Create the Operations topic automatically:
node scripts/telegram-topics.js \ --config ~/.openclaw/openclaw.json \ --chat <USER_CHAT_ID> \ --name "Operations" - Write the resulting thread ID into
shared/operations/channel-map.md - Update
cron/jobs.json— replace{{OPERATIONS_CHANNEL}}with the actual Operations channel address (format:chatId:threadId, e.g.,123456789:7)
If Group+Topics:
- Check if the user already has a supergroup:
- If not: guide them to create one (Create Group → toggle "Topics" on)
- 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"
- Agent reads the incoming message context to extract:
- Group chat ID from
{{To}} - Agent writes the chat ID into the channel config
- Group chat ID from
- Create the Operations topic automatically:
node scripts/telegram-topics.js \ --config ~/.openclaw/openclaw.json \ --chat <GROUP_CHAT_ID> \ --name "Operations" - Write the resulting thread ID into
shared/operations/channel-map.md - Update
cron/jobs.json— replace{{OPERATIONS_CHANNEL}}with the actual Operations channel address (format:chatId:threadId, e.g.,-100XXXXXXXXXX:7)
If DM-simple:
- "Send any message to your bot in Telegram"
- Agent reads the chat ID from the incoming message context
- Write chat ID into channel config — done
If Group-simple:
- Guide: "Add the bot to your group and send a message"
- Agent reads the group chat ID from the incoming message context
- Write chat ID into channel config — done
Step 5: Instance Setup + First Brand
After scaffolding and Telegram configuration, run the sub-skills:
-
Instance Setup (
instance-setupskill)- Owner name and timezone
- Communication language (owner-facing)
- Default content language
- Bot identity (name, emoji, personality)
- Updates:
shared/INSTANCE.md,workspace/IDENTITY.md
-
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.jsto create a topic named after the brand - The script returns the thread ID
- Agent writes the thread ID into
shared/operations/channel-map.mdand the brand config
- Agent calls
- For simple modes: no topic needed, skip thread ID
- Creates: brand profile, content guidelines, domain knowledge file, asset directories
Step 6: Verification + Gateway Restart
-
Restart gateway:
openclaw gateway restart -
Run diagnostics:
openclaw doctorThis validates: agent config, DM allowlist inheritance, session health, model availability, and workspace integrity.
-
Additional checks:
- [ ] Leader responds to messages
- [ ]
sessions_sendto 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-setupsub-skill, which guides you through installation and configuration.
Suggested first tasks after setup:
- Fill in your brand profile:
shared/brands/{brand_id}/profile.md - Test content creation: "Write a Facebook post for {brand}"
- Add more brands: "Add a new brand"
- 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_sendwhen 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 totasks/archive/. - Stale task detection is handled by a cron job (
stale-task-check, every 10 minutes) that scanstasks/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
| Document | Purpose | When to Read |
|---|---|---|
references/architecture.md |
Star topology, session model, parallelism | Understanding system design |
references/agent-roles.md |
Detailed agent capabilities and restrictions | Customizing team composition |
references/signals-protocol.md |
Complete signal dictionary | Debugging agent communication |
references/memory-system.md |
3-layer memory + knowledge capture | Understanding memory behavior |
references/approval-workflow.md |
Approval pipeline + owner shortcuts | Content publishing workflow |
references/troubleshooting.md |
Known issues (IPv6, etc.) + solutions | When something breaks |
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
| Script | Purpose | When to Run |
|---|---|---|
scripts/scaffold.sh |
Create directories, copy templates, set up symlinks | During initial setup |
scripts/patch-config.js |
Merge agent config into openclaw.json | During initial setup |
scripts/telegram-topics.js |
Create forum topics in Telegram DM or supergroup | During setup and when adding brands |
Sub-Skills
| Skill | Purpose |
|---|---|
instance-setup |
Configure owner info, language, bot identity |
brand-manager |
Add, edit, archive brands |
qmd-setup |
Install and configure QMD semantic search memory (optional) |
04 同仓库更多 Skills
crowd-prompting
A marketplace where AI agents improve prompts, system instructions, tool descriptions, and other text-based content with domain expertise from real-world operations — and earn tokens for valuable contributions.
ceaser
Interact with the Ceaser privacy protocol on Base L2 using the ceaser-mcp MCP tools. This skill uses the ceaser-mcp npm package for ALL operations -- shield, unshield, note management, and protocol queries. All ceaser tool calls use CLI subcommands (npx -y ceaser-mcp <subcommand>). Ceaser is a privacy-preserving ETH wrapper using Noir/UltraHonk zero-knowledge proofs on Base L2 (chain ID 8453).
ceaser-send
Fully automated private ETH transfer via Ceaser Protocol on Base L2 using the ceaser-mcp MCP tools. This skill uses the ceaser-mcp npm package for ALL operations -- all ceaser tool calls use CLI subcommands (npx -y ceaser-mcp <subcommand>). WARNING -- creates observable on-chain links between user wallet and hot wallet (see Privacy Warning). Generates an ephemeral hot wallet with BIP-39 mnemonic recovery, user funds it, agent signs and broadcasts the Shield TX automatically. Exactly one manual step required (funding the hot wallet with ETH). Uses Noir/UltraHonk zero-knowledge proofs.
dilemm-ai
Compete in DilemmAI, the prisoner's dilemma AI arena at dilemm.ai. Use when an OpenClaw agent wants to create an account, design and submit strategy prompts for their bot, enter matchmaking, analyze historical matches and leaderboards, and iterate on bot strategy. Requires headless browser automation (no API available yet).
moldium
Post and manage content on the Moldium blog platform. Triggered by "post to Moldium", "write a blog post", "publish an article", etc.
arxiv-paper-reviews
Interact with arXiv Crawler API to fetch papers, read reviews, submit comments, search papers, and import papers. Use when working with arXiv papers, fetching paper lists by date/category/interest, viewing paper details with comments, submitting paper reviews, searching papers by title, or importing papers from arXiv URLs via API at http://weakaccept.top:8000/.