memory-management

Memory Management - Claude Project Memory Management Skills

Skill Overview

Memory Management is a cross-session project memory system that uses a three-tier storage architecture (HOT/WARM/COLD) to automatically manage contextual data for SEO and GEO projects, helping Claude maintain continuity across different sessions.

Use Cases

1. Long-Term SEO Project Tracking

When you work on SEO optimization continuously over several weeks, you need to remember target keywords, competitor data, technical decisions, and unfinished tasks. This skill automatically saves key findings from the current session to the hot-cache, then loads them automatically in the next session, ensuring analysis remains coherent.

2. Knowledge Sharing in Collaborative Teams

Team members can quickly understand a project’s history, terminology definitions, and key decisions through the shared memory system, avoiding repeated communication and missed important information.

3. Regular Reporting and Progress Reviews

When you need to generate monthly SEO reports or review project progress, the skill can aggregate all stored historical data—including ranking changes, competitor analysis, and published content—to quickly produce a project status summary.

Core Features

Three-Tier Memory Architecture

HOT Cache (limit: 80 lines): Automatically loaded core project data, including target keywords, primary competitors, current priorities, and active activities

WARM Storage: Date-archived discovery data such as keyword research, competitor analysis, SERP snapshots, etc., loaded on demand

COLD Archive: Historical data not referenced for more than 90 days is automatically archived to save storage space; when needed, you can search manually

Automatic Memory Lifecycle Management

The skill automatically adjusts the “temperature” of data based on how frequently it’s cited: frequently used data is upgraded from WARM to HOT, while long-unused data is automatically downgraded and archived. No manual organization is required, and memory stays up to date.

Cross-Skill Data Aggregation

When the user asks, “What do we know right now?”, the skill automatically searches all memory subdirectories (keyword research, competitor analysis, content audits, etc.) and aggregates them to generate a complete project status report.

Frequently Asked Questions

Can Claude remember the last project conversation?

Yes. Memory Management automatically loads hot-cache.md at the start of each session via the SessionStart hook, including the keywords, competitors, priorities, and decisions you saved last time. At the end of the session, the Stop hook prompts you to save new findings.

How much space will the memory data take up?

The HOT cache is limited to within 80 lines, about 25KB. WARM and COLD storage use separate files named by date (e.g., 2024-03-15-keywords.md), which does not affect the main session context. Files not referenced for over 90 days are automatically moved to the archive directory.

How do I clean up old memory data?

The skill provides an automatic cleanup mechanism: HOT project data not referenced for 30 days is automatically downgraded to WARM, and WARM files not referenced for 90 days are automatically archived to COLD. You can also manually run the “Clean up stale data” command; the skill will list archivable files for confirmation.