skills.rest
ollieb89
ollieb89COMMUNITY
Explore all skills by this author →
GitHub

memory-curator

Deduplicate and prune AI agent memory entries using semantic embeddings.

Updated Mar 5, 2026
Search Tags:#curation#embeddings#agent#memory#deduplication#ollama
One-click install
npx skills add https://github.com/ollieb89/claw_imperium --skill memory-curator-ollieb89
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: memory-curator
Source: https://github.com/ollieb89/claw_imperium/tree/main/workspace/skills/memory-curator
Command: npx skills add https://github.com/ollieb89/claw_imperium --skill memory-curator-ollieb89

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires ollama, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill streamlines the management of an AI agent's long-term memory by automatically identifying and organizing redundant, outdated, or valuable information, ensuring the agent's knowledge base remains efficient and accurate.

Core Features & Use Cases

  • Semantic Deduplication: Identifies and suggests merging or linking of similar learning entries using advanced embedding models.
  • Memory Promotion: Helps promote durable rules and user preferences from temporary learning logs to permanent memory files like AGENTS.md or MEMORY.md.
  • Pruning Stale Memory: Assists in identifying and proposing the removal of obsolete or redundant information.
  • Use Case: Automatically review your agent's .learnings/ directory, find entries that describe the same git workflow, and suggest merging them to keep the memory concise.

Quick Start

Use the memory curator skill to deduplicate entries in your .learnings/LEARNINGS.md file with a similarity threshold of 0.92.

Frequently Asked Questions about memory-curator

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I deduplicate AI agent memory logs using semantic embeddings?▼

You can deduplicate AI agent memory by applying semantic embeddings to identify similar learning entries and suggest merging them. This process uses the nomic-embed-text-v2-moe model to analyze semantic similarity and streamline your knowledge base.

What is the best way to prune stale data from an AI agent's long-term memory?▼

Pruning stale agent memory involves identifying obsolete or redundant information and proposing its removal. This curation process keeps the knowledge base efficient by automatically reviewing logs and discarding data that is no longer relevant.

How do I promote temporary agent learnings to permanent memory files?▼

Promoting temporary learnings to permanent memory files like AGENTS.md or MEMORY.md involves identifying durable rules and user preferences. The curation process refines temporary logs and elevates valuable information to durable storage.

Does the memory curation process require Ollama to run locally?▼

Yes, Ollama is a required dependency for this memory curation process. It runs locally to support the nomic-embed-text-v2-moe model, which generates the semantic embeddings necessary for similarity analysis and deduplication.

How do I configure the similarity threshold for merging similar learning entries?▼

You configure the deduplication process by setting a specific similarity threshold, such as 0.92, when reviewing your LEARNINGS.md file. Entries meeting this similarity score are flagged for merging to keep memory concise.

Related Skills

Explore More Software Engineering Skills→
quazfentonquazfenton

cli-anything-ollama

Manage local Ollama models, generate text, and create embeddings via CLI.

Community
Intermediate
miriamdongmiriamdong
2

consolidate-memory

Merge duplicate memory entries and prune outdated facts from MEMORY.md.

Community
Intermediate
stevengonsalvezstevengonsalvez

Self-Improving Agent

Captures corrections and reflections into a structured, tiered memory hierarchy for future behavior improvement.

Community
Advanced
danmestasdanmestas
4

landing

Automate post-merge cleanup of worktrees, branches, and stale refs.

Community
Intermediate