メニュー
Deploy an AI chatbot in a Twitch channel that responds to a configurable trigger word (e.g. !bot), engages viewers with AI replies, and requires no Twitch API key — uses pure WebSocket IRC.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | twitch-ai-chatbot |
| description | Deploy an AI chatbot in a Twitch channel that responds to a configurable trigger word (e.g. !bot), engages viewers with AI replies, and requires no Twitch API key — uses pure WebSocket IRC. |
Connects to Twitch IRC via WebSocket and responds to a trigger word using an Anthropic model. No Twitch API key required for basic operation — only a bot account OAuth token.
Requirements: python3, aiohttp, websockets, Anthropic API key.
Use a separate Twitch account for the bot. Get a chat OAuth token:
https://twitchapps.com/tmi/oauth:XXXXXXXXXX tokencd ~/ares-skills/twitch-ai-chatbot/scripts
cp config.example.json config.json
Edit config.json:
{
"channel": "yourchannelname",
"trigger_word": "!bot",
"bot_login": "your_bot_account",
"oauth_token": "oauth:XXXXXXXXXX",
"model": "claude-haiku-4-5",
"anthropic_api_key": "sk-ant-XXXXXXXX",
"system_prompt": "You are a helpful and fun AI chatbot in a Twitch chat. Keep responses short (under 150 chars), casual, and engaging. No offensive content.",
"cooldown_seconds": 30,
"max_messages_per_30s": 18,
"log_file": "chatbot.log",
"storage_dir": "./storage"
}
Key fields:
channel — Twitch channel to join (lowercase)trigger_word — command viewers type to talk to the bot (e.g. !bot, !ai, !ask)bot_login — Twitch username of the bot accountoauth_token — OAuth token from twitchapps.com/tmi (include the oauth: prefix)model — Anthropic model ID. Recommended: claude-haiku-4-5 (fast + cheap)system_prompt — customize the bot's personality and rulescooldown_seconds — minimum seconds between bot responses (prevents spam)pip3 install aiohttp websockets
python3 ~/ares-skills/twitch-ai-chatbot/scripts/chatbot.py
Or with a custom config path:
python3 chatbot.py --config /path/to/config.json
All key config values can be set via env vars (take precedence over config.json):
export TWITCH_CHANNEL=yourchannelname
export TWITCH_LOGIN=your_bot_account
export TWITCH_OAUTH=oauth:XXXXXXXXXX
export TRIGGER_WORD=!bot
export AI_MODEL=claude-haiku-4-5
export ANTHROPIC_API_KEY=sk-ant-XXXXXXXX
export SYSTEM_PROMPT="You are a helpful bot..."
python3 chatbot.py
wss://irc-ws.chat.twitch.tv:443 — no API key neededtrigger_word: calls Anthropic API with last 10 chat messages as context@username <AI response>storage/chat_history.jsonlUpdate trigger_word in config.json. Examples: !ai, !ask, !gpt, !help
Update system_prompt. Example for a gaming bot:
You are an expert VALORANT coach in Twitch chat. Give quick tips and hype up plays. Max 150 chars per reply.
claude-haiku-4-5claude-sonnet-4-5tmux new-session -d -s twitch-bot 'python3 ~/ares-skills/twitch-ai-chatbot/scripts/chatbot.py'
Create ~/Library/LaunchAgents/com.twitch-ai-chatbot.plist with appropriate configuration and launchctl load it.
Bot connects but doesn't respond: Check trigger_word matches exactly. Verify bot account is in the channel (it joins automatically on connect).
Authentication failed: Regenerate OAuth token at twitchapps.com/tmi. Tokens expire.
Rate limited by Twitch: Increase cooldown_seconds. Default 30s is safe for most use.
AI replies are too long: Add "Max 150 characters" to your system_prompt.
Use when a user wants to translate text to or from any Indian language — Hindi, Tamil, Telugu, Bengali, Marathi, Kannada, Malayalam, Gujarati, Punjabi, Urdu, Odia, Assamese, Sanskrit — or says "translate this message", "translate this to Hindi", "what does this say in Tamil", "convert to Marathi", or similar. Also handles global languages (Spanish, French, Arabic, Chinese, Japanese, Korean, etc.). Uses Gemini 2.5 Flash. Also use when asked to detect what language a piece of text is written in.
Phrase-triggered emergency stop system for AI agents. Instantly halts all agent automation when a trigger phrase is detected in session logs, or via manual flag. Use when setting up a safety kill switch, emergency stop, or pause-all-automation mechanism for autonomous agents. Survives restarts via LaunchAgent watchdog.
The CRICD framework for structuring sub-agent task prompts — Context, Relevance, Instruction, Constraints, Demonstration. Produces dramatically better output from any LLM sub-agent. Use when spawning sub-agents, writing task prompts, or structuring complex instructions for AI models.
Formal protocol for draft → critique → revise loops in AI agent workflows. Defines when to use critique loops (money, public publishing, strategy) and when to skip (simple edits, routine coding). Prevents both over-engineering simple tasks and under-reviewing critical ones.
Monitor OpenClaw Gateway health by tailing logs for error patterns — 429 rate limits, auth failures, delivery failures, timeouts. Alert when error rate crosses thresholds. Use when setting up production monitoring for an OpenClaw Gateway deployment, or when a user says "watch the gateway", "monitor gateway errors", "alert me on gateway failures".
A 5-tier dispatch framework for coding tasks in AI agent workflows. Routes tasks from SIMPLE (one-file typo fix) to FULL (multi-day feature build) with appropriate rigor at each level. Prevents over-engineering simple tasks and under-planning complex ones.