Praxile - Governed local experience layer for coding agents

Coding agents can fix code, run tools, and complete tasks. But most of what they learn during a run disappears when the run ends. Praxile adds a governed experience layer to each repository: record how an agent run unfolded; extract evidence from trajectories, tool results, and user feedback; turn repeated lessons into reviewable proposals; store only approved experience under .praxile/; retrieve useful local experience in future runs. Praxile is not a model trainer, a hidden memory system, or a fully autonomous agent. It is an agent harness for making repository-specific experience safe to keep and easy to reuse.