Show HN: Autoharness – a self-learning, maintaining skill layer for Claude Code (github.com)

Autoharness, developed by Tigerless Labs, is a self-learning skill layer for Claude Code that autonomously manages and refines skills based on real user interactions. This innovative layer differentiates itself by merging similar skills, maintaining a clean repository that adapts to user needs without accumulating unnecessary duplicates. It achieves an impressive performance boost, elevating a baseline accuracy from 42% to 78% on CORE-Bench (HAL), showcasing significant advancements in machine learning operations by allowing skills to organically evolve and self-validate through actual usage, rather than relying solely on external benchmarks. The significance of Autoharness lies in its ability to streamline the skill-building process and enhance the usability of AI systems by reducing the need for separate data collection or manual intervention. Its architecture features components like CAP (capture), REF (reflect), and MNG (lifecycle) that work together to ensure that only relevant skills are retained and optimized. This not only minimizes overhead but also provides a pathway for continuous improvement in dynamic environments, setting a precedent for future AI models that aim for self-sustainability and efficiency without the dependence on extensive manual updates or rigorous benchmark evaluations.