continuous-learning-v2

You can extract reusable patterns from development sessions by using automated PreToolUse and PostToolUse hooks to observe activities, then running /learn to extract patterns and /checkpoint to save the state locally.

Confidence scoring for learned instincts is a mechanism that weights observed patterns based on their reliability, allowing high-confidence instincts to promote into reusable skills, rules, or agents.

Hook-based observation for continuous learning works by attaching PreToolUse and PostToolUse observers to development sessions, detecting opportunities to learn and apply patterns automatically as they emerge.

Learned instincts are stored locally at the ~/.claude/instincts/ directory, maintaining a structured lifecycle as they evolve from simple observations into promoted skills and agents.

Yes, you can promote observed development patterns into skills, rules, and agents. Clusters of instincts automatically evolve and promote as their confidence scores grow through repeated observations.

No external dependencies are required to automate learning from development sessions. The skill operates independently using local storage and deterministic hook-based observation without external packages.