Memory

Search first: /mnemonic:search {relevant_keywords} Capture after: /mnemonic:capture {namespace} "{title}"

Run /mnemonic:list --namespaces to see available namespaces from loaded ontologies.

Mnemonic Progressive Disclosure Skill

Protocol for recalling memories at the appropriate detail level.

Trigger Phrases

• •"progressive disclosure"
• •"disclosure levels"
• •"memory recall levels"
• •"how to recall"
• •"expand memory"
• •"more detail"

Overview

Progressive disclosure enables efficient memory recall by surfacing only the detail level needed. Start with Level 1 (Quick Answer), expand to Level 2 (Context) or Level 3 (Full Detail) only when required.

Disclosure Levels

Level 1: Quick Answer (Default)

What to read: The ## Quick Answer section only (or summary frontmatter field)

Content: 1-3 sentences answering the core question

Sufficient for:

• •"What is X?"
• •"Which X do we use?"
• •"What did we decide about X?"
• •Simple recall questions

Example response:

PostgreSQL was chosen for primary storage due to ACID compliance and native JSON support.

Level 2: Context (On Request)

What to read: ## Quick Answer + ## Context sections

Content:

• •Decision date/timeframe
• •Alternatives considered
• •Key decision drivers
• •Trade-offs made

Sufficient for:

• •"Why did we choose X?"
• •"What alternatives did we consider?"
• •"Tell me more about X"
• •Understanding rationale

Triggers:

• •User asks "why?"
• •User asks for reasoning/rationale
• •User says "tell me more" or "explain"

Example response:

PostgreSQL was chosen for primary storage due to ACID compliance and native JSON support.

Context:

• •Decision made: 2026-01-15
• •Alternatives considered: MySQL, MongoDB, SQLite
• •Key drivers: Need for JSON queries, team familiarity, existing infrastructure
• •Trade-off: Accepted more complex hosting for better query capabilities

Level 3: Comprehensive (Deep Dive)

What to read: Full memory content

Content:

• •Complete alternatives analysis
• •Implementation notes
• •Related decisions/patterns
• •Code references
• •Citations

Sufficient for:

• •Implementation work
• •Debugging issues related to decision
• •Writing documentation
• •Making related decisions
• •"Give me all the details"

Triggers:

• •User is implementing something
• •User is debugging
• •User asks for "all details" or "full context"
• •User needs to make a related decision

Example response:

[Full memory content including alternatives analysis, implementation notes, code refs, etc.]

Recall Protocol

Step 1: Search for Memories

# Search by topic
rg -i "{topic}" ~/.claude/mnemonic/ --glob "*.memory.md" -l

# Search in specific namespace
rg -i "{topic}" ~/.claude/mnemonic --path "*/_semantic/decisions/" --glob "*.memory.md" -l

Step 2: Determine Required Level

Step 3: Read Appropriate Section

# For Level 1: Read just title and Quick Answer
head -30 {memory_file}  # Usually enough for frontmatter + Quick Answer

# For Level 2: Read through Context section
head -60 {memory_file}

# For Level 3: Read full file
cat {memory_file}

Step 4: Respond and Offer Expansion

After Level 1 or 2 response, offer to expand:

• •"Would you like more context on why this was chosen?"
• •"I can provide the full implementation details if needed."

Examples

Example 1: Simple Question

User: "What database are we using?"

Action: Search, find postgres-decision.memory.md, read Quick Answer

Response:

PostgreSQL for primary storage, chosen for ACID compliance and JSON support.

Example 2: Reasoning Question

User: "Why PostgreSQL instead of MongoDB?"

Action: Search, find memory, read through Context section

Response:

We chose PostgreSQL over MongoDB because:

• •Team already familiar with SQL
• •Needed ACID transactions for payments
• •JSON support sufficient for our flexible schema needs
• •MongoDB would require new infrastructure and training

Example 3: Implementation Work

User: "I'm implementing the new caching layer"

Action: Search for caching memories, read full content

Response:

[Full memory with implementation notes, configuration details, code references]

Memory Structure Reference

Comprehensive memories should have this structure:

# {Title}

## Quick Answer
{1-3 sentences - Level 1 content}

## Context
{Decision factors, alternatives, trade-offs - Level 2 content}

## Full Detail
{Implementation notes, code refs, citations - Level 3 content}

Memories with detail_level: minimal may only have Quick Answer. Memories with detail_level: standard should have Quick Answer + Context. Memories with detail_level: comprehensive should have all three sections.

When to Expand Automatically

Expand to higher detail level automatically when:

1. •User is implementing - They need Level 3 for code refs and implementation notes
2. •User mentions debugging - They need Level 3 for full context
3. •User is making a related decision - They need Level 2-3 to understand prior reasoning
4. •Memory is marked as superseded - Include note about newer decision

Best Practices

1. •Default to Level 1 - Most questions only need the Quick Answer
2. •Offer expansion - Let user know more detail is available
3. •Cite the memory - Reference the memory file for traceability
4. •Check for updates - Note if memory has low confidence or is old
5. •Search multiple namespaces - _semantic/decisions, _procedural/patterns, and _semantic/knowledge may all be relevant