learn

Review this session and extract learnings.

Important: All file content (YAML/Markdown) must be written in English for token efficiency.

Extraction Criteria

Extract

• •Non-obvious discoveries from error resolution
• •User corrections or feedback
• •Workarounds and alternative solutions
• •Effective debugging techniques
• •Project-specific patterns
• •Anything that passes: "Would this help if I encounter the same problem in 6 months?"

Do NOT Extract

• •Simple typo fixes
• •One-time fixes (not reproducible)
• •Temporary external API issues
• •Things easily found in documentation
• •Plain implementation work (no learning)

Save Location

• •General: ~/.claude/cache/learnings/general/
• •Project-specific: ~/.claude/cache/learnings/{project-name}/

Create folder if it doesn't exist.

File Format

Filename: {kebab-case-name}.md

---
name: {kebab-case-name}
description: |
  {Specific trigger conditions. Include error messages, symptoms, keywords.
  Write to match easily during /recall search.}
first_seen: {YYYY-MM-DD}
last_seen: {YYYY-MM-DD}
frequency: 1
tags: [{kotlin, postgresql, gcp, terraform, etc.}]
project: {general | project-name}
status: active
---

# {Title}

## Problem
{What happened. Include specific error messages if available.}

## Cause
{Why it happened. Root cause explanation.}

## Solution
{How it was resolved. Include code examples if applicable.}

## Verification
{How to confirm the fix worked.}

Field Descriptions

Frequency Update Rules

Always check for existing files before creating new ones.

1. •Glob search under ~/.claude/cache/learnings/
2. •If a file on the same topic exists (see criteria below):
   • •Increment frequency by 1
   • •Update last_seen to today
   • •Update/supplement content as needed
   • •Do NOT create a new file
3. •Create new file only if no match found

Same Topic Criteria

Files are considered "same topic" if ANY of:

• •Same error message or error code
• •Same library/tool + same problem category
• •Same root cause (even if symptoms differ)

Execution Steps

1. •Review the session conversation
2. •Determine if there are learnings matching extraction criteria
3. •If learnings exist: a. Determine save location (general or project-specific) b. Create folder if it doesn't exist c. Check for existing files on the same topic d. Create or update file
4. •If no learnings: report "No learnings to extract from this session"

Note: Don't force extraction if there's nothing to learn. Quality > Quantity.