continuous-learning-v2

To turn Claude Code sessions into reusable knowledge, this Skill captures session observations and extracts them into atomic, confidence-scored instincts. It then evolves these instincts into higher-level capabilities like commands and agents.

To prevent cross-project contamination of agent memory, you can use project-scoped instincts alongside global patterns. This ensures observations and learned behaviors remain isolated to their specific project contexts.

The deterministic evolution pipeline assigns confidence scores to session observations as it captures them. These scored instincts are then stored in a local persistence store before evolving into skills or agents.

Yes, you need YAML frontmatter containing name and description fields to extract instincts from Claude sessions. This frontmatter anchors the automated pipeline that processes and scores the observations.

Yes, you can evolve atomic instincts into standalone commands and agents. The local memory pipeline uses confidence scoring to determine when an instinct is ready to scale into a higher-level tool.

Using a local persistence store for agent memory limits instinct sharing to a single environment. This approach prioritizes deterministic, reproducible results over distributed or cloud-based memory synchronization.