LiveClaw Live Streaming Executor

Problem

When an OpenClaw Agent runs a task, external channels may only see the final text instead of the observable execution process. LiveClaw targets live-streaming scenarios where the Agent's instructions, reasoning, tool calls, and results are pushed to a real-time dashboard and delivered through TRTC as a publicly accessible viewer URL. It also supports TTS status announcements and viewer-side message delivery, making it useful for presenting an agent live execution stream as live content.

How It Works

• Initialization: It reads or generates $WORK_DIR/config.json, using TRTC SDKAppID and SecretKey to create the room, RTMP URL, and a UserSig with a default 7-day validity.
• Streaming pipeline startup: It starts frame rendering, supervision, optional TTS worker, and the HTTP viewer service, and requires Stream ready or a sufficient wait before generating the viewer page.
• Event reporting: Once the live system is running, each Agent response must report task, info, result, or error events through agent_bus.py; plain-text replies also need segmented info events, otherwise the dashboard and TTS cannot show the full text.
• Two-way interaction: Viewers can trigger the Agent through IM messages, while the Agent's results are displayed through the stream, event feed, and TTS output.

Boundaries And Caveats

The skill depends on Tencent Cloud TRTC Experience Edition or above and reserves specific ports. SecretKey, CAM credentials, and callback tokens must not be written into logs or public repositories. After reinitializing, the viewer page must be regenerated to avoid the wrong room. It is primarily for live streaming and Agent status visualization, not for unrelated general-purpose tasks.