skills.rest
wangrenzhu-ola
wangrenzhu-olaCOMMUNITY
Explore all skills by this author →
GitHub

hkt-memory

Layer stored content from L2 to L1 to L0 with hybrid vector and BM25 retrieval.

2|1|Updated Apr 17, 2026
Search Tags:#recall#llm#mcp#retrieval#memory#hook#layering
One-click install
npx skills add https://github.com/wangrenzhu-ola/GaleHarnessCodingCLI --skill hkt-memory
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: hkt-memory
Source: https://github.com/wangrenzhu-ola/GaleHarnessCodingCLI/tree/main/vendor/hkt-memory
Command: npx skills add https://github.com/wangrenzhu-ola/GaleHarnessCodingCLI --skill hkt-memory

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires flask>=3.0.0, openai>=1.0.0, requests>=2.31.0, tqdm>=4.66.0, numpy>=1.24.0, and includes scripts (resource) components.

What problem does it solve?

HKT-Memory v5.0 delivers production-grade long-term memory management with automatic multi-layer storage, ensuring critical knowledge is preserved and accessible across teams and projects.

Core Features & Use Cases

  • Three-layer storage: L2 full content triggers L1 overviews and L0 abstracts for fast recall.
  • LLM-based structured extraction: titles, key points, decisions, actions, and timelines are captured automatically.
  • MCP protocol support and Claude Code hooks: enable seamless client integration and automated capture/recall.
  • Hybrid search: vector embeddings plus BM25 for robust recall across large knowledge bases.
  • Use cases include architecture logs, project retrospectives, and knowledge archives across teams.

Quick Start

Install and run memory via the CLI, then store a memory in layer all and retrieve with a query to verify layering and recall.

Frequently Asked Questions about hkt-memory

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

FAQPage Schema
How does automatic layering work for LLM long-term memory management?▼

Long-term memory management uses automatic layering to cascade stored content from L2 full text to L1 overviews and L0 abstracts. This ensures fast recall while preserving detailed knowledge across large software development project bases.

How do I integrate persistent memory capture with Claude Code hooks and MCP?▼

You can achieve persistent memory capture using MCP protocol support and Claude Code hooks. This enables seamless client integration and automated capture and recall within your software development workflows.

What is the best way to retrieve architecture logs using vector and BM25 hybrid search?▼

The best way to retrieve architecture logs is using a hybrid retrieval system combining vector embeddings and BM25. This approach ensures robust recall across large knowledge bases by matching semantic meaning and exact keywords.

Does HKT-Memory work without an available LLM API key?▼

Yes, HKT-Memory provides robust fallbacks when API keys are unavailable. You can still store and retrieve memories, although LLM-based structured extraction for titles and key points will be limited without a key.

How do I store project retrospectives and extract decisions automatically?▼

You can store project retrospectives via the CLI, which triggers LLM-based structured extraction. This automatically captures titles, key points, decisions, actions, and timelines from the stored content across memory layers.

Related Skills

Explore More Software Engineering Skills→
InternScienceInternScience
28

molclaw-scp-server

Configure and manage the MCP server for MolClaw tool communication.

Official
Intermediate
arbgjrarbgjr

session-analyzer

Parse Claude Code session logs to extract decisions, blockers, and learnings.

Community
Intermediate
masermediagroup-stackmasermediagroup-stack
1

using-vector-databases

Connect vector databases with embedding models for semantic search and RAG systems.

Community
Advanced
yonatangrossyonatangross
3

pgvector-search

Combine vector embeddings with BM25 keyword search using Reciprocal Rank Fusion.

Community
Advanced