AI Skill

OpenClaw Skill — Local AI Assistant Integration

OpenClaw is an open-source local AI assistant that runs on your computer (Mac/Linux/Windows) and operates through chat apps like WhatsApp, Telegram, Discord, and Slack. This skill — `codivupload-social-manager` on ClawHub — plugs CodivUpload directly into OpenClaw's marketplace. Your local AI agent gains the ability to schedule, publish, and analyze social media content across YouTube, Instagram, X, Facebook, TikTok, Threads, and Pinterest (7+ launched, with Bluesky / LinkedIn / Snapchat / GBP in active rollout) — all while your data stays on your own machine.

7+ launched — YouTube, Instagram, X (Twitter), Facebook, TikTok, Threads, Pinterest. Plus Bluesky, LinkedIn, Snapchat, and Google Business Profile in active rollout.

Best for:Privacy-focused indie developers, local-first AI users, founders who want their personal AI assistant to handle social posting without sending data to a SaaS scheduler.

Install MCP serverAll skills

What this skill teaches the LLM

The skill is a markdown file the LLM reads when activating. Below is the focused context this skill bakes in — so Claude/ChatGPT/Cursor doesn't have to learn from API errors.

  • OpenClaw skill format: YAML frontmatter (name, description, metadata.openclaw.os, requires.bins, requires.config) + markdown body for agent instructions
  • Skill installs to ~/.openclaw/workspace/skills/codivupload/SKILL.md
  • Activates on triggers: 'CodivUpload' brand mention, 'cross-post', 'schedule social media via API', 11-platform name + automation/API intent
  • Config requirement: CODIVUPLOAD_API_KEY (set via openclaw config set)
  • Prefers MCP server (codivupload-mcp) over direct API when available — fewer agent tokens used
  • Falls back to curl/SDK code generation when MCP not configured — agent uses OpenClaw's exec tool
  • Compatible with all OpenClaw-supported LLMs: Claude 3.5+, GPT-4o+, local models with tool-use
  • Published to ClawHub marketplace under tags: social-media, scheduling, api, automation, mcp, publishing
  • Open source — fork on GitHub at github.com/Codivion/CodivUpload, skill files in npmjs/openclaw-skill/

Example prompts

These are real prompts you can give an AI client with this skill installed. The skill provides the LLM with the right tool calls, parameter names, and platform-specific quirks so the response is correct on the first try.

1

“Schedule this video to post on TikTok, Instagram, and YouTube tomorrow at 9am.”

What the LLM does

OpenClaw activates the codivupload skill, calls MCP tool create_post (or generates curl request) with profile_name + platforms array + scheduled_at, confirms success across all 3 platforms.

2

“Cross-post my latest blog announcement to LinkedIn and X.”

What the LLM does

Agent reads recent blog post, generates platform-tailored captions (long-form for LinkedIn, threaded for X), uses MCP cross-post action with platform-specific override params.

3

“What's the best time to post for my TikTok audience?”

What the LLM does

Skill instructs agent to call analytics endpoint, return per-platform best_time data, present human-readable summary in chat (Telegram/Discord/Slack).

4

“Set up a 24/7 YouTube live stream with this MP4 source.”

What the LLM does

Agent guides through BYOP setup if needed, then triggers create_live_stream with source URL, confirms FFmpeg relay started, monitors via watchdog.

Install in 2 steps

The skill itself is a markdown file. The MCP server is the runtime. Install both for the full experience.

  1. 1Configure MCP server in your AI client

    Paste this configuration into Claude Desktop / Cursor / Zed's MCP config file. Replace cdv_your_api_key with your real key from pricing.

    # Drop SKILL.md into your OpenClaw skills workspace
mkdir -p ~/.openclaw/workspace/skills/codivupload
curl -L https://raw.githubusercontent.com/Codivion/CodivUpload/main/npmjs/openclaw-skill/SKILL.md \
  -o ~/.openclaw/workspace/skills/codivupload/SKILL.md

# Set your CodivUpload API key
openclaw config set CODIVUPLOAD_API_KEY=cdv_your_api_key

# Optional: install the MCP server (gives the agent direct tool access)
npm install -g codivupload-mcp

# OR install via ClawHub
clawhub install codivupload

  2. 2Add the skill markdown

    Download npmjs/skills/openclaw/SKILL.md from the GitHub repo and place it under ~/.claude/skills/codivupload-openclaw/SKILL.md (Claude Desktop) or your project's .claude/skills/ directory.

    For ChatGPT: paste the SKILL.md content into a Custom GPT's Instructions field at gpts.openai.com/editor.

    github.com/Codivion

Frequently asked

Why install via OpenClaw instead of just using the MCP server directly with Claude Desktop?+

OpenClaw is local, private, and operates through chat apps you already use (WhatsApp, Telegram, Discord, Slack). With this skill, you can ask your phone via WhatsApp 'schedule this to TikTok at 9am' and it just happens — no separate dashboard, no copy-paste between apps. The skill gives OpenClaw's local agent CodivUpload-specific knowledge so it doesn't hallucinate API parameters.

Does this work with my existing CodivUpload account or do I need a new one?+

Same account. The skill uses your existing CODIVUPLOAD_API_KEY (find it in dashboard → Settings → API Keys). All 11 platforms, all features, all profile/workspace settings carry over. Free plan included (10 uploads/month).

Is OpenClaw really running locally, not sending my data to a server?+

Yes — OpenClaw is open source and runs on your machine. The CodivUpload skill is just markdown instructions that teach OpenClaw's local LLM how to use the CodivUpload API. The actual API calls go from your machine to api.codivupload.com directly. No third-party intermediary.

Can I publish my own version of this skill to ClawHub?+

Yes — the skill is MIT-licensed. Fork the repo, customize the frontmatter (e.g., narrow to specific platforms), bump version, run 'clawhub skill publish'. Your fork lives alongside ours in the marketplace.

Does the skill work with local LLMs like Llama or Mistral?+

Yes if the local LLM supports tool-use / function calling. The skill is markdown-based instructions; any LLM that can read instructions and decide to call exec/MCP tools will work. Quality may vary — Claude 3.5 + GPT-4o are most reliable; Llama-70B and DeepSeek work but with more retries.

Related skills

CodivUpload Generic Skill

Generic AI skill for Claude Desktop, Claude Code, ChatGPT Pro, Cursor, Zed. Covers all 11 social platforms via MCP server. Anthropic Skills format, MIT licensed.

Claude Agency + Workspace Skill

AI skill for agency operations — multi-tenant workspaces, branded team invites, role-based access (Admin/Editor/Viewer), profile move, billing cascade. For Claude, ChatGPT, Cursor.

Ready to give your AI agent posting tools?

Free CodivUpload plan covers 10 uploads/month. Skills are MIT licensed.

Get an API key