dsh-record-replay

As AI agents increasingly venture into everyday desktop environments, teaching them to accurately understand and replicate complex, interactive human workflows remains a significant hurdle. Traditional recording methods or simple key-mapping scripts often struggle with dynamic graphical user interfaces and contextual dependencies, leading to fragile automations that break easily. The dsh-record-replay plugin addresses this gap by bringing first-class observation-and-learning capabilities to DeepSeek Harness. It empowers agents to master software interaction logic much like humans do—by watching and learning from real screen interactions rather than relying solely on brittle code.

Built around the open-record-replay macOS toolchain, the plugin orchestrates a complete capture-to-deploy pipeline using six specialized tools. These span initial accessibility verification, live session recording, event stream extraction, structural validation, and final skill packaging. When a user demonstrates a workflow on their Mac, the plugin captures the raw interaction events and immediately checks them against strict contracts to ensure data integrity. The validated evidence is then transformed into standardized SKILL.md packages. It seamlessly integrates with existing native skill creators while offering a robust fallback generator that builds progressive-disclosure documentation and evaluation datasets automatically, with hot-reloadable configuration support for rapid iteration.

Designed primarily for DeepSeek Harness developers, local AI researchers, and workflow automation engineers, this plugin serves as a bridge between human expertise and machine execution. Whether you aim to convert repetitive GUI operations into reliable, reusable skills or want to experiment with cutting-edge paradigms of LLM-system interaction, dsh-record-replay provides the essential infrastructure. With strictly typed outputs and built-in validation, teams can efficiently transform operator intuition into durable agent capabilities, significantly reducing the development cycle and maintenance overhead for sophisticated desktop automation projects.