m2wise_skill

Overview

This skill provides a Memory-to-Wisdom engine that gives an AI agent long-term memory and the ability to evolve higher-level guidance from stored memories. It captures user facts, preferences, and commitments, retrieves relevant context before responses, and consolidates raw memories into reusable wisdom over time. The goal is safer, more personalized, and progressively smarter agent behavior.

How this skill works

The skill exposes an SDK and optional MCP tools to add conversational messages as memories, query a user's contextual memories when composing replies, and run background consolidation routines. Online calls (add_message, get_context) handle real-time storage and retrieval. Periodic background jobs (trigger_sleep, trigger_dream) group memories into wisdom drafts and verify them into principles, schemas, or skills.

When to use it

• When a user states a preference or interaction rule (e.g., tone, format, content restrictions).
• When a user shares personal facts or commitments you should recall later.
• Before answering complex or personalized questions to fetch relevant context.
• After finishing a major task or session to consolidate new patterns into wisdom.
• When you want to expose or audit accumulated user guidance or behavioral rules.

Best practices

• Proactively save strong preferences or commitments without waiting for explicit "remember" requests.
• Always query retrieved context for ambiguous or personalization-sensitive replies.
• Run consolidation (sleep and dream) periodically to convert repeated memories into stable wisdom.
• Treat memories by type (preference, fact, commitment) and prefer recent, high-confidence items during retrieval.
• Use the MCP tools if your runtime supports Model Context Protocol for lower-latency operations.

Example use cases

• Remembering user formatting and tone preferences to apply across future answers.
• Storing and recalling personal profile facts to personalize recommendations and reminders.
• Consolidating repeated interaction patterns into a principle like "avoid emojis when user prefers concise prose."
• Running background consolidation at the end of a workday to surface new skills the agent can apply.
• Searching archived memories to resolve a user dispute about a previously stated commitment.