ClawVoice

**Judgement:** ClawVoice is a focused, practical skill for adding real-time voice I/O to OpenClaw agents via a WebSocket-backed Python client. It appears mature enough for development and demo use, and likely suitable for many production voice-interaction scenarios where basic reliability and responsiveness are acceptable. **What it does well:** - Bridges a live voice session to the agent using simple CLI commands (`send`, `recv`, `listen`, `agent`). - Supports both single-turn and streaming, multi-turn listening with configurable timeouts. - Provides an `agent` loop mode that continuously forwards user messages to `openclaw agent --agent main` and returns stdout to the user. - Uses a straightforward JSON protocol with clear message types and behavior guidelines. **Key risks / limitations:** - Each interaction is mediated via a Python script (`uv run python client.py`), which may introduce overhead and complicate very high-throughput or low-latency workloads. - Reliability depends on the external WebSocket voice server; the documentation does not describe reconnection or robust error recovery beyond simple timeouts and exit codes. - Blocking behavior (`recv`, `listen`, `agent`) requires careful orchestration to avoid deadlocks or unintended parallelism in more complex agent workflows. - No inherent guardrails on conversational content; safety depends almost entirely on the agent logic using the skill. **Recommended scenarios:** - Quickly enabling voice for an existing text-based OpenClaw agent (especially for demos, pilots, and internal tools). - Real-time conversational assistants where moderate latency is acceptable and traffic volume is not extreme. - Prototyping or experimenting with voice UX flows using the `listen` and `agent` modes to simulate live sessions. - Development environments where a local WebSocket voice server (default `ws://localhost:3111/connect`) is available and controllable.