memory

name	memory
description	Access persistent memory across Claude sessions. Search past conversations, recall decisions, store key insights. Use when needing context from previous work or to save important information for future sessions.
allowed-tools	Bash, Read
metadata	{"version":"1.0.0","requires-secrets":[],"author":"Cognova","repository":"","installed-from":""}

Memory Skill

Access and manage persistent memory from previous Claude sessions. Memories are automatically extracted from conversations and can be explicitly stored.

Memory Types

Type	Icon	Description
decision	[D]	Architectural or design decisions made
fact	[F]	Key facts or information learned
solution	[S]	Solutions to problems encountered
pattern	[P]	Patterns or conventions discovered
preference	[*]	User or project preferences
summary	[~]	Summarized context from sessions

Commands

Search memories

python3 ~/.claude/skills/memory/memory.py search "<query>" [options]

Options:

• --type <type> - Filter by memory type
• --project <path> - Filter by project path
• --limit <n> - Max results (default: 10)

Examples:

python3 ~/.claude/skills/memory/memory.py search "authentication"
python3 ~/.claude/skills/memory/memory.py search "database" --type decision
python3 ~/.claude/skills/memory/memory.py search "API" --project "/Users/me/project"

Get recent memories

python3 ~/.claude/skills/memory/memory.py recent [limit] [options]

Options:

• --type <type> - Filter by memory type

Examples:

python3 ~/.claude/skills/memory/memory.py recent
python3 ~/.claude/skills/memory/memory.py recent 5
python3 ~/.claude/skills/memory/memory.py recent --type decision

Store a memory

python3 ~/.claude/skills/memory/memory.py store "<content>" [options]

Options:

• --type <type> - Memory type (default: fact)
• --project <path> - Associated project path
• --relevance <0-1> - Relevance score (default: 0.9)

Examples:

python3 ~/.claude/skills/memory/memory.py store "We use PostgreSQL with Drizzle ORM"
python3 ~/.claude/skills/memory/memory.py store "API rate limit set to 100 req/min" --type decision
python3 ~/.claude/skills/memory/memory.py store "User prefers pnpm over npm" --type preference

List decisions

python3 ~/.claude/skills/memory/memory.py decisions [options]

Options:

• --project <path> - Filter by project
• --limit <n> - Max results (default: 20)

Examples:

python3 ~/.claude/skills/memory/memory.py decisions
python3 ~/.claude/skills/memory/memory.py decisions --project "cognova"

Find memories about a topic

python3 ~/.claude/skills/memory/memory.py about "<topic>" [options]

Options:

• --limit <n> - Max results (default: 15)

Examples:

python3 ~/.claude/skills/memory/memory.py about "error handling"
python3 ~/.claude/skills/memory/memory.py about "testing strategy"

Preview session context

python3 ~/.claude/skills/memory/memory.py context [options]

Shows what context would be injected into a new session.

Options:

• --project <path> - Project path
• --limit <n> - Max memories (default: 5)

Natural Language Patterns

When users say things like:

• "What did we decide about..." -> Use search or decisions
• "Remember that we..." -> Use store
• "What do you know about..." -> Use about
• "Show recent context" -> Use recent or context
• "Save this insight..." -> Use store

How Memory Works

Automatic Extraction

Memories are automatically extracted from conversations:

1. PreCompact Hook - Before context compaction, key insights are extracted
2. Stop Hook - After Claude responds, important facts are saved asynchronously

Session Context

When a new session starts, relevant memories are automatically injected based on:

• Project path matching
• Recency
• Relevance score
• Access frequency

Relevance Scoring

Each memory has a relevance score (0-1):

• Newly stored memories start at 0.9
• Frequently accessed memories maintain higher scores
• Old, unused memories gradually decay

Best Practices

1. Store key decisions - When making architectural choices, use store --type decision
2. Check before changes - Before major refactors, use about to check relevant history
3. Review decisions - Use decisions to see past architectural choices
4. Explicit storage - For critical insights, explicitly store rather than relying on auto-extraction