live-evo-memory

To stop Claude Code from repeating mistakes, use memory-guided task execution to store failure experiences. It captures lessons after each task and retrieves them via embedding similarity before running new non-trivial actions to ensure past errors are avoided.

Embedding-based retrieval for coding guidelines ranks past task experiences by semantic similarity adjusted by historical weight. This mechanism surfaces high-signal lessons, enabling Claude Code to synthesize 3-5 actionable guidelines before executing a new task.

You manage a dynamic memory store using developer tools like memory_manager.py. It supports init, add, search, update-weight, feedback, prune, decay, and export operations to maintain and evolve the knowledge base of task experiences.

Yes, Claude Code supports an always-on lifecycle workflow through retrieve, execute, capture, and feedback phases. This lifecycle initializes memory per project, applies weighted experiences to current tasks, and refines future guidance by creating failure experiences.

Yes, you need OpenAI to use this embedding-based memory retrieval system. The Skill explicitly depends on the OpenAI library to generate embeddings and calculate semantic similarity for ranking past task experiences and guidelines.

The best way to apply memory-guided guidelines is initializing a project memory store first. Before every non-trivial task, the system automatically retrieves relevant weighted experiences and synthesizes 3-5 actionable guidelines to direct Claude Code's actions.