mem-search

Search persistent cowork-mem notes using FTS5 keyword and TF-IDF semantic matching.

8|1|Updated Mar 21, 2026

Search Tags:#search #memory #semantic-search #tf-idf #knowledge-retrieval #fts5

Ready to run

npx skills add https://github.com/MSApps-Mobile/claude-plugins --skill mem-search-msapps-mobile

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: mem-search
Source: https://github.com/MSApps-Mobile/claude-plugins/tree/main/plugins/cowork-mem/skills/mem-search
Command: npx skills add https://github.com/MSApps-Mobile/claude-plugins --skill mem-search-msapps-mobile

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Quickly find and retrieve relevant information from persistent cowork-mem so users and agents can recall past decisions, errors, notes, and contextual details without manual searching through logs or documents.

Core Features & Use Cases

Dual-mode search: Fast, exact keyword search via FTS5 and broader semantic retrieval via TF-IDF vectors to catch paraphrases and related concepts.
Filtered queries: Narrow results by type (decision, error, insight, file_edit, tool_use, note, summary) to focus on the most relevant memories.
Fetch full context: Retrieve full observation records by id for reading or synthesis before taking action.
Use Case: Ask whether a past deployment error was documented and what mitigation steps were taken, then fetch the original note to confirm details.

Quick Start

Search memory for notes about the authentication bug and summarize the key decisions and fixes found.

Frequently Asked Questions about mem-search

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

FAQPage Schema

How do I search past decisions and notes using semantic matching and keyword lookup?▼

Search past decisions and notes using dual-mode retrieval that combines FTS5 keyword lookup for exact matches and TF-IDF vector semantic search to catch paraphrases. This returns relevant records from persistent memory without manual log searching.

What is the best way to retrieve context about a previous deployment error from memory?▼

Retrieve deployment error context by querying memory with semantic search, then fetch the full observation record by id. This confirms original mitigation details and provides complete contextual information for synthesis before taking action.

Can I filter memory search results by type to focus on specific past events?▼

Filter memory search results by type including decision, error, insight, file_edit, tool_use, note, and summary. This narrows retrieval to focus on the most relevant memories matching your specific query intent.

Does semantic search work with paraphrased queries when exact keywords are not in the notes?▼

Semantic search works with paraphrased queries using TF-IDF vectors to catch related concepts even when exact keywords are absent. This broader retrieval complements FTS5 exact keyword matching for comprehensive memory coverage.

How do I fetch the full observation record after finding a relevant note in search results?▼

Fetch the full observation record by using its id after an initial search. This allows reading complete details and synthesizing information before taking action, ensuring accurate context retrieval from persistent memory.

What are the limitations of using FTS5 keyword search alone for retrieving past notes?▼

FTS5 keyword search alone misses paraphrases and related concepts that TF-IDF semantic search captures. Relying solely on exact keyword matching limits memory retrieval coverage when query phrasing differs from stored note content.