Analyzes coding sessions to detect and codify your preferences and corrections into durable instructions.
AutoSkill serves as a persistent learning mechanism for Claude, bridging the gap between one-off chat corrections and permanent project documentation. By analyzing session history for signals like repeated patterns, explicit corrections, and approvals, it identifies high-confidence improvements for your Skill files and CLAUDE.md. This ensures that Claude evolves with your codebase, remembering project-specific architecture, naming conventions, and utility preferences without requiring constant manual instruction in every new session.
Key Features
0140 GitHub stars
02Proposes minimal, reversible edits to Skill files and CLAUDE.md for review
03Uses a confidence-scoring system to filter noise from actionable rules
04Distinguishes between skill-specific behavior and project-wide conventions
05Detects durable coding preferences from chat corrections and approvals
06Supports manual review and git-based rollback for all suggested changes
Use Cases
01Capturing and enforcing team-specific error handling and naming conventions
02Updating library preferences, such as switching between utility functions like cn() or clsx()
03Codifying project-specific architectural rules after initial setup corrections
