Memory Management Skill

This skill teaches you how to use the ccmemory persistent context system.

Overview

The ccmemory plugin provides persistent memory across Claude Code sessions via:

1. Semantic memory - ChromaDB vector database for storing/retrieving facts
2. Session handoff - .ccmemory/session.md for working memory
3. Doc index - .ccmemory/doc-index.md for project documentation inventory

MANDATORY: Capture Triggers

You MUST store information to memory when ANY of these patterns occur:

1. User Corrections (Highest Priority)

Recognize these patterns and IMMEDIATELY store the correction:

• "No, that's wrong..." / "That's not how it works..."
• "I already told you..." / "We discussed this..."
• "Stop doing X" / "Don't assume Y"
• "The actual way is..." / "It's actually..."
• Any frustrated tone about repeated mistakes
• Any time the user explains something you got wrong

Response to corrections:

1. Acknowledge briefly (no over-apologizing)
2. IMMEDIATELY store the correct information via memory service
3. State: "Stored to memory: [brief description]"
4. Continue with the corrected approach

2. Project Facts

Store when the user explains:

• How a subsystem or component works
• Project-specific patterns or conventions
• Why something is done a certain way
• Gotchas, edge cases, or "watch out for X"
• Technology choices and their rationale
• File organization or naming conventions

3. Decisions

Store when the user says:

• "I've decided..." / "I'm going to..." / "I chose to..."
• "Let's use X instead of Y"
• "The approach will be..."
• Any commitment after weighing options

4. Preferences

Store when the user expresses:

• Code style preferences
• Tool or library preferences
• Workflow preferences
• "I prefer..." / "Always do X" / "Never do Y"

How to Store Facts

Use the memory service tools to store facts. Each fact should include:

• What: The specific information
• Context: Why it matters or when it applies
• Source: Where this came from (user statement, discovered during task)

Example storage pattern:

Store: "Database migrations must use alembic, not raw SQL"
Context: "Project convention for schema changes"

Session Protocols

Session Start

1. Read .ccmemory/session.md for previous session context
2. Query memory service for facts related to the current task
3. Check doc-index for relevant documentation

Session End

Update .ccmemory/session.md with:

• Key learnings from this session
• Decisions made (with reasoning)
• Open questions
• Files modified

Documentation Index

The .ccmemory/doc-index.md lists project documentation. Keep it updated when:

• New documentation files are created
• You discover undocumented but important files
• Documentation structure changes

Memory vs Session.md vs Docs

Information Type	Where to Store
Permanent project facts	Memory service
Session-specific context	session.md
Detailed documentation	Project doc files
Quick reference facts	Memory service
Corrections/gotchas	Memory service (highest priority)

Self-Check

After each user message, ask yourself:

1. Did the user correct me? → Store immediately
2. Did the user explain something about the project? → Store it
3. Did the user make a decision? → Store it
4. Did the user express a preference? → Store it

If yes to any: use the memory service before continuing with the task.