Autoharness is a self-learning skill layer for Claude Code that automatically captures, consolidates, and maintains reusable skills from your coding sessions without requiring manual configuration or separate benchmarking infrastructure.
The tool addresses the problem that Claude Code's capabilities depend on a manually-maintained skill layer that becomes stale and bloated over time. Autoharness solves this by learning skills directly from real work you're already doing. When a session accumulates enough productive interactions, the tool distills those interactions into a skill and stores it in your local skills directory. Rather than simply accumulating skills, it compares new episodes against existing ones and merges skills that handle the same scenarios, preventing duplicate or near-duplicate entries. Skills survive based on actual usage in subsequent sessions rather than held-out benchmark scores, and the tool only modifies skills it created itself, leaving your hand-written skills and other plugins untouched.
You should adopt this tool if you use Claude Code regularly and want to build up a personalized, self-maintaining skill library without manual curation. It works best for teams or individuals who generate enough diverse coding sessions to benefit from skill consolidation. The tool requires only Python on your system with no external dependencies and zero configuration to start—it begins watching your sessions immediately after installation. The `/learn` command lets you manually distill a lesson from your current session when you want to capture something immediately rather than waiting for the background process.
The project shows consistent activity with regular updates to its skill learning and merging logic. The tool maintains detailed per-skill ledgers recording the scenario and decision behind each skill creation or update, providing a foundation for building benchmarks from real usage patterns. The codebase demonstrates active refinement of the validation mechanisms that determine whether skills remain useful across sessions.
