agentskills.codes
ME

memory

by huivrotikiData & AnalyticsSource

Persists and retrieves project context for consistent AI agent behavior.

RAGAutomation

Install

mkdir -p .claude/skills/memory-huivrotiki && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/15786" && unzip -o skill.zip -d .claude/skills/memory-huivrotiki && rm skill.zip

Installs to .claude/skills/memory-huivrotiki

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Hierarchical memory — Supermemory RAG + ChromaDB persistence
60 charsno explicit “when” trigger
Intermediate

Key capabilities

  • →Retrieve context based on a query
  • →Store descriptions of solutions
  • →Persist project-specific information
  • →Manage user preferences
  • →Index knowledge for retrieval

How it works

The skill uses `smartRetrieve` for pre-flight lookup to get context for a task and `smartStore` for post-flight persistence to save solutions. It manages memory in scopes like user and project.

Inputs & outputs

You give it
A query string or a description of a solution
You get back
Retrieved context or stored memory

When to use memory

  • →Retrieving project history
  • →Storing architectural decisions
  • →Persisting user preferences

About this skill

Supermemory RAG — Иерархическая память

Pre-flight Lookup

Перед началом любой задачи:

const { smartRetrieve } = await import('./scripts/memory-mcp.js');
const context = await smartRetrieve("описание задачи", 3);

Или через skill:

Вызови memory-retrieve skill с query = описание задачи

Post-flight Persist

После фикса бага, рефакторинга или архитектурного решения:

const { smartStore } = await import('./scripts/memory-mcp.js');
await smartStore("описание решения", { source: "файл", tags: "тип", type: "fix" });

Scope разделения

ScopeЧто хранитьПример
userПредпочтения, стиль, привычки"Предпочитает Tailwind arbitrary values"
projectСпецифика репозитория, архитектура"Shadow Router использует Playwright CDP"

Memory Priority

  1. CLAUDE.md — всегда загружен (this file)
  2. handoff.md — загрузить в начале сессии
  3. SESSION.md — текущее состояние сессии
  4. SKILL.md — загрузить только когда нужен навык

Compaction

  • Threshold: 80% контекста использовано
  • После compaction: записать summary в SESSION.md
  • Приоритеты загрузки: CLAUDE.md → handoff.md → SESSION.md → SKILL.md

Embedding Safety (M1 8GB)

  • КРИТИЧНО: Всегда keep_alive: 0 после эмбеддингов (nomic-embed-text)
  • Никогда Promise.all для множественных эмбеддингов — только for...of
  • Модель nomic-embed-text занимает ~280MB VRAM — выгружай сразу

Indexing

source .venv/bin/activate && python scripts/index_knowledge.py

ChromaDB Config

  • Path: ./memory/shadow_memory (PersistentClient, на диск)
  • Embedding: nomic-embed-text через Ollama REST API
  • Chunks: 500 chars с 50-char overlap

When not to use it

  • →When multiple embeddings are needed simultaneously without sequential processing
  • →When `keep_alive: 0` is not used after embeddings

Prerequisites

Ollama REST APIpythonscripts/memory-mcp.jsscripts/index_knowledge.py

Limitations

  • →Embedding model `nomic-embed-text` occupies ~280MB VRAM
  • →Embeddings should not use `Promise.all` for multiple operations
  • →Context compaction occurs at 80% usage threshold

How it compares

This skill provides hierarchical memory with persistence using ChromaDB, allowing agents to retain information across sessions, unlike a manual approach that requires re-entering context.

Compared to similar skills

memory side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
memory (this skill)04moReviewIntermediate
dust-call-agent14moNo flagsIntermediate
mcp-index05moNo flagsBeginner
llm-ops03moNo flagsAdvanced

Try saying

Example prompts that trigger this skill in your AI assistant.

You might also like

dust-call-agent

dust-tt

Call a Dust agent to get information (read a slack thread, a notion URL, a google drive document...), perform an action (post a message to slack, create a calendar event, ...), provide context on any topic regarding Dust (the company, current discussions, customers...) or in general have the Dust agent perform a given task.

10

mcp-index

ms-mfg-community

Invoke BEFORE complex reasoning, library doc lookups, cross-session memory, or multi-agent coordination to discover available MCP tools.

00

llm-ops

Anhvu1107

ALWAYS use this when the request matches LLM OPS: LLM Operations -- RAG, embeddings, vector databases, fine-tuning, prompt engineering avancado, custos de LLM, evals de qualidade e arquiteturas de IA para producao.

00

moai-mcp-builder

jg-chalk-io

Enterprise MCP (Model Context Protocol) server development using FastMCP

00

data-engineering

pluginagentmarketplace

ETL pipelines, Apache Spark, data warehousing, and big data processing. Use for building data pipelines, processing large datasets, or data infrastructure.

13192

firecrawl-scraper

jackspace

Scrape and extract web content, convert HTML to markdown, and bypass bot protection for dynamic sites using Firecrawl API.

24149

Search skills

Search the agent skills registry