Memory Palace Skill

Transform information into unforgettable memories using the ancient method of loci combined with the Red Queen adversarial learning protocol.

Commands

Quick Start

/memory-palace create "TypeScript Mastery" "Ancient Library"
/memory-palace store "generics"
/memory-palace recall
/memory-palace red-queen weak-spots

Storage Location

Memory is organized into global and project-specific contexts:

~/memory/
├── config.json              # System configuration
├── global/                  # Cross-project knowledge
│   ├── palace-registry.json
│   ├── meta-index.md
│   ├── learning-journal.md
│   └── *.json               # Global palaces (system-design, algorithms, etc.)
│
└── project/
    └── {projectId}/         # Project-specific knowledge
        ├── palace-registry.json
        ├── meta-index.md
        └── *.json           # Project palaces (codebase, domain, conventions)

Context Detection

Project ID is derived from: hash(git_remote_url) or hash(folder_path)

Command Implementation

/memory-palace create <name> [theme]

1. If no theme provided, suggest 3 options:

   • “Ancient Library” (hierarchical knowledge)
   • “Space Station” (technical systems)
   • “Journey Path” (sequential learning)

2. Create palace JSON structure:

{
  "name": "<name>",
  "created": "YYYY-MM-DD",
  "theme": "<theme>",
  "activeLocus": "entrance",
  "loci": [
    {
      "id": "entrance",
      "name": "Grand Entrance",
      "anchor": "<vivid anchor description>",
      "memories": [],
      "children": [],
      "parent": null
    }
  ]
}

3. Save to ~/memory/global/<slugified-name>.json (or project context)
4. Update palace-registry.json in current context
5. Describe the entrance vividly to anchor it

/memory-palace store <topic>

1. Ensure palace is active (or ask which to use)

2. Ask what information to memorize

3. Transform using SMASHIN SCOPE:

   • Substitute abstract → concrete
   • Movement – make it animated
   • Absurd – impossible/exaggerated
   • Sensory – all 5 senses
   • Humor – make it funny
   • Interact – user is part of scene
   • Numbers – encode with shapes
   • Symbols – visual puns
   • Color – vivid, unusual
   • Oversize – dramatic scale
   • Position – precise placement
   • Emotion – strong feelings

4. Present image to user, refine if needed

5. Place at specific locus

6. Update palace JSON and meta-index

7. Brief “walk back” to reinforce

/memory-palace recall [topic]

1. If topic specified, search palace for matching memories

2. If no topic, offer options:

   • Full tour (walk all loci)
   • Recent (last 5 memories)
   • Random quiz

3. For each memory:

   • Describe the locus vividly
   • Present the image
   • Reveal the content
   • Ask if user wants elaboration

/memory-palace red-queen [strategy]

Strategies:

• random – Random sampling (default)
• weak-spots – Focus on low-confidence items
• depth-first – Complete palace tour
• cross-link – Questions connecting concepts
• adversarial – Edge cases and failure modes

Protocol:

1. Load palace and meta-index
2. Launch Examiner agent (haiku) to generate questions
3. Launch Learner agent (haiku) for blind recall
4. Launch Evaluator agent (haiku) to score
5. Update learning journal with gaps
6. Optionally launch Evolver to strengthen weak memories

/memory-palace status

Display:

• Total palaces
• Total memories
• Last accessed
• Weak spots count
• Next review due (spaced repetition)

Palace JSON Schema

{
  "name": "string",
  "created": "YYYY-MM-DD",
  "theme": "string",
  "description": "string",
  "activeLocus": "locus-id",
  "loci": [
    {
      "id": "unique-id",
      "name": "Location Name",
      "anchor": "memorable feature",
      "description": "detailed scene",
      "memories": [
        {
          "id": "memory-id",
          "subject": "topic",
          "image": "vivid SMASHIN SCOPE image",
          "content": "actual information",
          "created": "YYYY-MM-DD",
          "confidence": 1-5,
          "lastRecalled": "YYYY-MM-DD",
          "recallCount": 0,
          "linkedTo": ["other-memory-ids"]
        }
      ],
      "children": ["child-locus-ids"],
      "parent": "parent-id or null"
    }
  ]
}

Red Queen Protocol

“It takes all the running you can do, to keep in the same place.” — The Red Queen, Through the Looking-Glass (Lewis Carroll, 1871)

Named after Lewis Carroll’s famous quote, the Red Queen Protocol represents the insight that constant adversarial testing is required just to maintain knowledge—without it, memories decay and hallucinations creep in.

Key Results: -37% retrievals needed, +23% retention for weak memories, F1=0.92 hallucination detection

Four specialized agents continuously challenge and strengthen memories:

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│  EXAMINER   │────►│   LEARNER   │────►│  EVALUATOR  │
│  (haiku)    │     │   (haiku)   │     │   (haiku)   │
│ Generate Qs │     │ Blind recall│     │ Score gaps  │
└─────────────┘     └─────────────┘     └──────┬──────┘
                                               │
                                               ▼
                                        ┌─────────────┐
                                        │   EVOLVER   │
                                        │   (opus)    │
                                        │ Strengthen  │
                                        └─────────────┘

Agent Roles

Evolution Triggers

1. Manual: /memory-palace red-queen [strategy]
2. Contextual: When discussing a topic with stored memories
3. Scheduled: Based on spaced repetition intervals

Context-Aware Evolution

• Global memories evolve from general discussions (~/memory/global/)
• Project memories evolve from codebase-specific work (~/memory/project/{id}/)
• Each context has its own learning journal tracking gaps

Survival Structure

For context loss recovery:

1. meta-index.md contains short anchors that trigger full recall
2. palace-registry.json tracks all palaces for chunked loading
3. Loading strategy:
   • Minimal: meta-index only
   • Standard: + registry
   • Full: + all palace JSONs

Navigation

• Table of Contents
• References – Techniques and templates
• Examples – Sample palaces
• Palaces – Your memory palaces
• Red Queen Protocol – Adversarial testing details

Best Practices

1. Start small: 5-7 loci per palace initially
2. Review regularly: Run red-queen weekly
3. Connect palaces: Link related memories across palaces
4. Personalize: More personal = stronger memory
5. All senses: Sound, smell, texture, taste, not just visual
6. Trust the absurd: Impossible images stick better