memory-taxonomy

name	memory-taxonomy
description	General domain memory classification — subject/predicate taxonomy, permanence levels, tagging strategy, and example facts
version	1.0.0

Memory Taxonomy — General Butler

This skill defines the classification framework for storing and retrieving freeform facts in the General butler's memory layer. Use it whenever you call memory_store_fact to ensure consistent, discoverable, and well-prioritized memory entries.

General Domain Taxonomy

The General butler handles catch-all data that does not fit specialist domains (health, finance, travel, education, etc.). Use flexible subject/predicate structures.

Subject

The subject anchors the fact to a named entity, concept, or topic:

• Personal facts: Use "user" for facts about the owner
• Project/concept: Use the project or concept name ("project-alpha", "rust-programming")
• Place or resource: Use the place or resource name ("coffee-shops", "vacation-planning")

Examples: "user", "project-alpha", "rust-programming", "vacation-planning", "coffee-shops"

Predicates

Predicate	When to use
goal	Personal or project goals
preference	User preferences not covered by a specialist butler
resource	Useful links, articles, or tools
idea	Brainstorming notes, future plans
note	General observations or reminders
deadline	Time-sensitive tasks or dates
status	Current state of a project or activity
recommendation	Recommendations (places, books, tools)

Permanence Levels

Level	When to use
stable	Long-term preferences, recurring patterns unlikely to change
standard	Most general facts — current state that may change over weeks/months (default)
volatile	Temporary notes, time-sensitive reminders, one-off tasks

Tags

Use tags for cross-cutting organization. Good defaults:

urgent, learning, work, personal, someday-maybe, places, action-required

Example Facts

# From: "I want to learn Rust this year"
memory_store_fact(
    subject="rust-programming",
    predicate="goal",
    content="learn Rust programming language in 2026",
    permanence="standard",
    importance=6.0,
    tags=["learning", "programming", "2026-goals"]
)

# From: "Good coffee shop: Blue Bottle on 5th St"
memory_store_fact(
    subject="coffee-shops",
    predicate="recommendation",
    content="Blue Bottle on 5th St - good coffee",
    permanence="standard",
    importance=4.0,
    tags=["places", "coffee", "local"]
)

# From: "Password reset link expires in 24 hours"
memory_store_fact(
    subject="password-reset",
    predicate="deadline",
    content="password reset link expires in 24 hours",
    permanence="volatile",
    importance=7.0,
    tags=["urgent", "action-required"]
)

Question Answering Flow

When the user asks a question:

1. Search memory first: memory_search(query=<question>) or memory_recall(topic=<subject>)
2. Search entities: item_search() with relevant query terms
3. Combine sources: Synthesize information from memory and entity storage
4. Respond: notify(channel=<channel>, message=<answer>, intent="reply", request_context=<ctx>)

Example:

User: "What was that coffee shop I liked?"
1. memory_search(query="coffee shop recommendation")
2. item_search(collection="places", query={"type": "coffee"})
3. Find: "Blue Bottle on 5th St"
4. notify(channel="telegram", message="Blue Bottle on 5th St — you saved that as a good coffee spot.",
          intent="reply", request_context=<from session>)

Extraction Philosophy

• Extract liberally — capture facts even from casual notes or tangential remarks
• Use standard by default — only use volatile for urgent/time-sensitive facts, stable for long-term preferences
• Tags enable discovery — choose tags that support finding facts across different future contexts
• Importance scale: 1–10. Urgency and personal significance raise importance; passing remarks lower it