ai-comm
Cross-AI CLI communication for Kitty terminal. Use when collaborating with other AI assistants (Codex, Gemini, Aider, Cursor, OpenCode), delegating tasks, or requesting code review.
When & Why to Use This Skill
The ai-comm skill enables advanced Cross-AI CLI communication within the Kitty terminal, allowing different AI assistants to collaborate, delegate tasks, and interact in real-time. It is a powerful orchestration tool for developers using multiple AI agents like Codex, Gemini, Aider, and Cursor, facilitating a multi-agent ecosystem directly from the command line.
Use Cases
- Multi-Agent Code Review: Delegate specific code segments to specialized models (e.g., Gemini or Codex) for a second opinion or deep security analysis without leaving your current session.
- Architectural Validation: Query a secondary AI assistant in a parallel Kitty window to verify system design decisions and explore alternative architectural patterns.
- Task Delegation: Distribute sub-tasks such as unit test generation or documentation to other running AI agents to maintain a high-velocity development workflow.
- Cross-Model Verification: Compare outputs from different AI models to reduce hallucinations and ensure the highest level of technical accuracy for complex implementations.
- Collaborative Debugging: Send terminal output and error logs to a separate AI instance to get fresh debugging perspectives while you continue working on the main codebase.
| name | ai-comm |
|---|---|
| description | Cross-AI CLI communication for Kitty terminal. Use when collaborating with other AI assistants (Codex, Gemini, Aider, Cursor, OpenCode), delegating tasks, or requesting code review. |
| allowed-tools | Bash, Read |
ai-comm
Cross-AI CLI communication tool for Kitty terminal. Enables AI assistants running in separate Kitty windows to communicate with each other.
Workflow
ai-comm list-ai-windows— Find available AI windowsai-comm send <MESSAGE> -w <ID>— Send message and get response
Important Notes
Replying to ai-comm messages. Just output your response as normal text—the sender automatically captures your terminal output. Do NOT use ai-comm to reply (causes deadlock). If you need the sender to act, include the request in your response.
For long responses, request file output. Bash tool has a 30000-character limit. If you expect a long response, ask the AI to write to a markdown file in the project directory (
/tmpand other external paths require manual approval on target AI — avoid them).Only use documented parameters. Never use parameters not listed in this SKILL or
ai-comm --help. Hidden/internal parameters exist for debugging only.
When to Use
- Delegate code review to Codex or Gemini
- Get second opinions on architecture decisions
- Request specialized analysis from another AI
- Verify implementations with alternative models
Resources
- For command reference, see reference.md
- For workflow examples, see examples.md