continuous-learning-v2

To stop Claude Code from forgetting coding patterns, you need a continuous learning mechanism that captures and retains learned workflows from interactive sessions. This Skill uses hooks to automatically observe tool use events and store project-specific coding instincts, preventing repetitive manual pattern re-entry across sessions.

Yes, you can isolate learned code patterns using project-scoped storage. This Skill stores atomic learned behaviors per-project by default, ensuring framework-specific patterns stay isolated to their respective git repositories while only universal best practices are shared globally across codebases.

Automatic session observation works by using hooks to capture tool use events in Claude Code interactive sessions. This mechanism detects user corrections, error resolutions, and repeated workflows without requiring manual input, storing them as confidence-weighted atomic instincts for future use.

The best way to turn repeated workflows into reusable skills is through instinct evolution and promotion. This Skill clusters high-confidence learned instincts into reusable skills, commands, or agents, automatically promoting them to a global scope when they appear across two or more projects with 80% or higher confidence.

No, manual triggering is not required. This Skill enables deterministic session observation via hooks, automatically capturing 100% of tool use events to detect user corrections, error resolutions, and repeated workflows without any manual input throughout the interactive session.

You should avoid using automated learning if your workflow requires manual curation of every coding pattern or if your projects lack distinct git-repository boundaries, because this Skill relies on project-scoped storage and automatic hook-based observation to isolate framework-specific patterns and prevent cross-project contamination.