Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
ai-collab v2.0.0 changelog - Major version release with formatting and metadata updates. - SKILL.md and _meta.json modified; content and core instructions unchanged. - No functional or protocol changes—existing users do not need to update configuration or scripts.
What This Skill Does
Runs two OpenClaw agents in parallel: a primary interactive Claude Code session and a background daemon triggered by filesystem inbox messages. Agents coordinate via a structured chat log with a tagged protocol that enforces closed loops. Supports Claude-Claude, Claude-GPT, and GPT-GPT pairings.
Structured tagging with mandatory DONE/BLOCKED closure prevents agents from leaving tasks open-ended, which is the main failure mode in ad-hoc multi-agent setups.
When to use it
- Fetching live crypto prices in the background while the primary agent handles user requests
- Offloading a multi-step Kraken API pull to the daemon while the main agent continues coding
- Routing Telegram messages from a group chat into the agent pipeline
- Having GPT review code that Claude just wrote, without switching tools
- Running a daemon watchdog that restarts itself if silent for more than 15 minutes
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: get the current BTC price and continue working
- 1Agent A writes [TASK:price-check] to the inbox using atomic mktemp + mv
- 2Daemon detects the new file via inotifywait and reads the message
- 3Agent B runs claude --print with the task context
- 4Agent B responds [DONE:price-check] with the price and logs it to chat.log
- 5poll_chatlog.sh routes the response back to Agent A via openclaw agent
Agent A receives '[AgentB]: [DONE:price-check] BTC $94,230 as of 03:15 UTC' and resumes its primary task
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.