Kibitz is a VS Code extension and CLI tool designed to provide real-time, decoded feeds of AI agent actions from sessions like Claude Code and Codex. It allows users to monitor multiple AI interactions simultaneously, understand their execution flow, and efficiently dispatch prompts across existing or new sessions. This tool is beneficial for developers, researchers, and power users who need enhanced visibility and control over complex AI agent workflows, enabling better coordination and debugging.

How It Works

Kibitz operates through a VS Code extension and a command-line interface (CLI). It monitors active Claude and Codex sessions, generating a live commentary feed that translates agent actions into human-readable insights. A key feature is its ability to dispatch prompts to any active watched session or initiate a new session on the current provider. The system provides explicit dispatch status events (queued, started, sent, failed) and supports provider-aware model handling, offering a structured approach to interacting with multiple AI agents.

Quick Start & Requirements

• Install:
  • VS Code Extension: Install from the VS Code Marketplace or search "Kibitz" in the Extensions panel.
  • CLI (macOS/Linux): brew install kibitzsh/tap/kibitz
  • CLI (npm): npm install -g @kibitzsh/kibitz
• Prerequisites: Node.js 20+, npm 10+, VS Code 1.85+, and at least one authenticated provider CLI (codex/codex.cmd or claude/claude.cmd).
• Links: VS Code Marketplace (search "Kibitz").

Highlighted Details

• Live commentary feed for Claude Code and Codex sessions.
• Cross-session prompt dispatch to existing or new sessions.
• Slash commands for session control (/help, /pause, /resume, /clear, /focus, /model, /preset, /interval, /update) and targeting (/1, /2, /target).
• Provider-aware model handling and strict dispatch status events.

Maintenance & Community

The project outlines a robust release flow (npm run cr) for publishing to VS Code Marketplace, npm, and Homebrew, including automated checks and version bumping. It also details setup for GitHub auto-releases and daily download digest emails, indicating active development and a structured maintenance process. No direct community links (e.g., Discord, Slack) are provided in the README.

Licensing & Compatibility

• License: MIT. This license is permissive and generally allows for commercial use and integration into closed-source projects.
• Compatibility: Supported on macOS and Windows for both VS Code panel and Terminal CLI. Linux CLI support is "best effort." Compatible with VS Code and Cursor.

Limitations & Caveats

Linux CLI support is explicitly noted as "best effort." Prompt targeting is limited to active sessions visible within the watcher window, and multi-target broadcast in a single send is not currently implemented.