Witness – a voice-anchored evidence recorder for AI coding agents (github.com)

Witness, a new voice-anchored evidence recorder for AI coding agents, has been launched to enhance the debugging process for developers. This innovative tool allows users to record their audio and the actions being taken within a managed browser environment simultaneously. For example, if a user says, "this button does nothing," Witness captures not only the audio but also related technical data such as clicks, console errors, and DOM changes, packaging this comprehensive evidence for AI coding agents to analyze. This capability significantly reduces the time and effort traditionally spent on documenting bugs, offering a streamlined, voice-centric approach to technical troubleshooting. The significance of Witness lies in its ability to provide a richer context for AI-driven coding agents, allowing them to understand the 'why' behind reported issues. By linking spoken descriptions directly with causal evidence in real-time, developers can efficiently convey complex problems without extensive documentation. Key features include live transcription, a variety of control commands (like marking elements and pausing recordings), and a focus on user privacy, ensuring sensitive information remains secure. Technical professionals can integrate the tool seamlessly with Node.js and preferred browsers, making it an agile addition to any developer's toolkit aimed at improving collaboration with AI agents.