/learn - Pattern Extraction for GROWI

Extract reusable problem-solving patterns from development sessions and save them as auto-invoked Skills.

Core Purpose

Capture "non-trivial problems" solved during GROWI development, converting them into reusable skills that will be automatically applied in future sessions.

Pattern Categories to Extract

Focus on four key areas:

1. Error Resolution — Document what went wrong, root causes, and fixes applicable to similar issues (e.g., Mongoose query pitfalls, Next.js hydration errors, TypeScript strict mode issues)

2. Debugging Techniques — Capture non-obvious diagnostic steps and tool combinations (e.g., MongoDB query profiling, React DevTools with Jotai, Vitest debugging patterns)

3. Workarounds — Record library quirks, API limitations, and version-specific solutions (e.g., @headless-tree edge cases, Socket.io reconnection handling, SWR cache invalidation)

4. GROWI Patterns — Note codebase conventions, architecture decisions, and integration approaches (e.g., feature-based structure, Jotai + Socket.io sync, API v3 design patterns)

Skill File Structure

Extracted patterns are saved in appropriate skill directories based on the scope of the pattern:

Workspace-specific patterns:

• apps/{workspace}/.claude/skills/learned/{topic-name}/SKILL.md
• packages/{package}/.claude/skills/learned/{topic-name}/SKILL.md
• Examples: patterns specific to a single app or package

Global patterns (monorepo-wide):

• .claude/skills/learned/{topic-name}/SKILL.md
• Examples: patterns applicable across all workspaces

File Template

---
name: descriptive-name
description: Brief description (auto-invoked when working on related code)
---

## Problem
[What was the issue]

## Solution
[How it was solved]

## Example
[Code snippet or scenario]

## When to Apply
[Specific conditions where this pattern is useful]

GROWI-Specific Examples

Topics commonly learned in GROWI development:

Apps/app-specific (apps/app/.claude/skills/learned/):

• page-save-origin-semantics — Origin-based conflict detection for collaborative editing
• socket-jotai-integration — Real-time state synchronization patterns
• api-v3-error-handling — RESTful API error response patterns
• mongodb-query-optimization — Mongoose indexing and aggregation patterns

Global monorepo patterns (.claude/skills/learned/):

• virtualized-tree-patterns — @headless-tree + @tanstack/react-virtual optimizations (if used across apps)
• jotai-atom-composition — Derived atoms and state composition (if shared pattern)
• turborepo-cache-invalidation — Build cache debugging techniques
• pnpm-workspace-dependencies — Workspace dependency resolution issues

Quality Guidelines

Extract:

• Patterns that will save time in future sessions
• Non-obvious solutions worth remembering
• Integration techniques between GROWI's tech stack
• Performance optimizations with measurable impact

Avoid:

• Trivial fixes (typos, syntax errors)
• One-time issues (service outages, environment-specific problems)
• Information already documented in existing Skills
• Feature-specific details (these stay in code comments)

Workflow

1. User triggers /learn after solving a complex problem
2. Review the session to identify valuable patterns
3. Draft skill file(s) with clear structure
4. Autonomously determine the appropriate directory:
   • Analyze the pattern's scope (which files/modules were involved)
   • If pattern is specific to a workspace in apps/* or packages/*:
     • Save to {workspace}/.claude/skills/learned/{topic-name}/SKILL.md
   • If pattern is applicable across multiple workspaces:
     • Save to .claude/skills/learned/{topic-name}/SKILL.md
5. Skills automatically apply in future sessions when working on related code

Directory Selection Logic

Workspace-specific (save to {workspace}/.claude/skills/learned/):

• Pattern involves workspace-specific concepts, models, or APIs
• References files primarily in one apps/* or packages/* directory
• Example: Page save logic in apps/app → apps/app/.claude/skills/learned/

Global (save to .claude/skills/learned/):

• Pattern applies across multiple workspaces
• Involves monorepo-wide tools (Turborepo, pnpm, Biome, Vitest)
• Shared coding patterns or architectural principles
• Example: Turborepo caching pitfall → .claude/skills/learned/

Learned skills are automatically invoked based on their description when working on related code.