bfly123/claude_code_bridge
Overview
This skill sends a user-provided message to a chosen AI provider (gemini, codex, opencode, droid) and ends the turn immediately so the backend can process asynchronously. It is designed for explicit delegation: use it only when the user asks you to forward their prompt to another AI provider. The skill minimizes token overhead and preserves persistent context across providers.
How this skill works
You call the skill with the provider name and pipe the message via stdin (heredoc or pipe) to avoid shell globbing. The command queues the request with optional foreground/background flags or environment overrides, prints a task ID and log path, and returns immediately. After invoking, the agent responds with "[Provider] processing..." and ends the turn; result retrieval is tracked separately.
When to use it
- When the user explicitly asks to send their prompt to a specific provider (gemini, codex, opencode, droid).
- When you need to forward a command or query to another AI without interpreting or answering it yourself.
- When you want persistent context and minimal token overhead across multi-AI workflows.
- When you need asynchronous processing and will check results or logs later.
Best practices
- Always pass the message via stdin (heredoc or pipe) rather than as CLI arguments to avoid shell globbing.
- Specify the provider as the first argument exactly: gemini, codex, opencode, or droid.
- Use --foreground or --background flags, or environment variables (CCB_ASK_FOREGROUND / CCB_ASK_BACKGROUND) to control execution mode.
- Set CCB_CALLER to indicate which agent is delegating (example: CCB_CALLER=codex) before calling ask.
- After running, rely on the returned task ID and log path to track progress; do not attempt to wait for the result in the same turn.
Example use cases
- Delegate a math or code question to Gemini: CCB_CALLER=codex ask gemini <<'EOF' ... EOF
- Send a long code-generation prompt to Codex in background: ask codex --background <<'EOF' ... EOF
- Route a debugging session to OpenCode or Droid for specialized handling and then monitor the task logs.
- Chain agents: set CCB_CALLER to one agent and forward a subtask to a different provider for parallel processing.
FAQ
No. Always provide the message via stdin (heredoc or pipe) to prevent shell globbing and ensure correct transmission.
How do I check if the request failed?
Use the provider-specific ping command (for example, ping gemini) and inspect the task ID and log file path printed when the request was accepted.
4 skills
This skill sends your request to a chosen AI provider via ask and ends the turn for seamless multi-agent collaboration.
This skill coordinates multi-CLI collaborative planning with Codex as coordinator to design solutions across Claude, Gemini, and OpenCode.
This skill lets you view the latest reply from a specified AI provider and optionally display multiple recent conversations.
This skill reports which CCB providers are mounted by validating session and daemon status, returning a concise JSON map.