Save Learnings

Save Session Learnings

Purpose

Review the current session and persist any valuable learnings, corrections, patterns, or project knowledge to the memory system. This ensures future sessions benefit from today's discoveries.

When to use

• End of a productive session with new discoveries
• After receiving user corrections or feedback
• After debugging a tricky issue
• When Stop hook reminds you to save learnings
• When user explicitly asks to save

Process

1. Read current memory state

• Read MEMORY.md index to see what already exists
• Glob memory files to understand current coverage

1. Identify what's new this session

• User corrections/feedback → feedback type memory
• Project patterns discovered → project type memory
• External resources/tools found → reference type memory
• User preferences/role info → user type memory

1. Check for updates vs new files

• If an existing memory covers the same topic, update it
• Only create a new file if the topic is genuinely new
• Keep memory file names descriptive: feedback_testing.md, project_api_patterns.md

1. Write memory files

• Use frontmatter format: name, description, type
• Keep content concise and actionable
• Focus on "what" + "how to apply" — not verbose narratives

1. Update MEMORY.md index

• Add one-line pointer for each new file
• Keep entries under 150 chars
• Keep total index under 200 lines

Output

List of memory files created or updated, with a one-line summary of each.

Quality bar

• Every saved memory has a clear "how to apply" — not just a fact dump
• No duplicate memories — always check existing files first
• MEMORY.md index stays clean and scannable
• Memory descriptions are specific enough to judge relevance in future sessions

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: enbiyagoral
• Source: enbiyagoral/.claude
• License: MIT

Install and usage instructions live in the source repository linked above.