Can it retrieve information from previous conversations?

Yes, it uses integrated search and recent-chat tools to pull context from prior sessions when you ask it to 'remember what we did before' or extract learnings from recent history.

Does it modify my source code or config files automatically?

No. The skill follows a strict 'Propose → Approve → Execute' workflow. It will always present a diff and ask for your explicit approval before making any changes to your files.

How does the Claude Memory Extractor store information?

The skill saves learnings into a tiered system of Markdown files: Layer 0 for global preferences (~/.claude/CLAUDE.md), Layer 1 for project-specific rules, and Layers 2-3 for evolving notes and deep reference topics.

What happens if my MEMORY.md file gets too long?

The skill includes a 'line check' feature. If a memory file exceeds 170 lines, it will warn you and suggest moving lower-priority sections to specific topic files to ensure the agent remains performant.

Claude Memory Extractor

Name: Claude Memory Extractor
Author: gupsammy

bygupsammy

•

生产力与工作流

Persists project insights, technical patterns, and behavioral preferences across sessions using a structured multi-layer memory hierarchy.

The extract-learnings skill enables Claude to proactively capture and organize technical insights, project-specific conventions, and universal behavioral preferences. By utilizing a tiered memory hierarchy—ranging from global settings to detailed topic-specific documentation—it ensures that valuable context discovered during development is preserved and automatically surfaced in future sessions. This prevents repetitive explanations, maintains architectural consistency, and builds a robust long-term knowledge base directly within your repository and global configuration.

主要功能

01Context retrieval tools to pull learnings from past chat histories

02Multi-layer memory management across Global, Project, and Session-specific files

03Automated distillation of complex conversations into concise, actionable insights

04Mandatory approval workflow to review and edit proposed changes before execution

0546 GitHub stars

06Semantic deduplication to prevent redundant entries in your documentation

使用场景

01Creating global behavioral rules that persist across all your development projects

02Saving non-obvious debugging patterns or 'gotchas' discovered during a fix

03Standardizing project conventions in CLAUDE.md for architectural consistency

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add gupsammy/claudest extract-learnings

For use in Claude.ai and ChatGPT

Download Skill