Skill

memory-management

Cross-project workplace memory backed by Basic Memory MCP. Decodes shorthand, acronyms, nicknames, and project codenames so Claude understands requests like a colleague would. Use when the user mentions a person, term, or project Claude does not know, says "remember this" or "X means Y", asks "who is X" or "what does X mean", or starts any non-trivial task that benefits from prior context.

npx claudepluginhub turek/cowork-global-memory --plugin global-memory

Tool Access

This skill uses the workspace's default tool permissions.

Preview

Cross-project memory that works in every Cowork session regardless of

SKILL.md

Similar Skills

cache-components

139.3k

Guides Next.js Cache Components and Partial Prerendering (PPR): 'use cache' directives, cacheLife(), cacheTag(), revalidateTag() for caching, invalidation, static/dynamic optimization. Auto-activates on cacheComponents: true.

cache-components

mcp-builder

124.2k

Guides building MCP servers enabling LLMs to interact with external services via tools. Covers best practices, TypeScript/Node (MCP SDK), Python (FastMCP).

9 files

anthropics-skills-13

Stats

Stars0

Forks0

Last CommitMay 3, 2026

Actions

View Source View Plugin View on GitHub View README

Help us improve

Share bugs, ideas, or general feedback.

Memory management (Basic Memory backed)

Cross-project memory that works in every Cowork session regardless of which folder is currently selected. Storage is plain markdown on disk. Access is via the basic-memory MCP server.

The goal

Decode shorthand into understanding. Without memory, "ask todd to do the PSR for oracle" is meaningless. With memory, Claude knows todd is Todd Martinez, PSR is Pipeline Status Report, oracle is the Oracle Systems deal.

Architecture

Two-tier system. Hot cache for fast lookup, full notes for depth.

Permalink	Purpose
`claude`	Hot cache. Top ~30 people, common acronyms, active projects. Read first.
`glossary`	Full decoder ring — every term, nickname, codename.
`preferences`	Communication and code preferences.
`index`	Map of the memory store.
`people/<slug>`	One file per person.
`projects/<slug>`	One file per project.
`context/<topic>`	Company, teams, stack.
`decisions/<YYYY-MM-DD>-<slug>`	ADR-style notes.
`daily/<YYYY-MM-DD>`	Append-only journal, one per day.

Lookup flow

For any non-trivial request, decode shorthand before acting:

Call mcp__basic-memory__read_note with permalink claude. Covers roughly 90% of decoding.
If a term wasn't in the hot cache, call mcp__basic-memory__read_note with permalink glossary.
If deeper context is needed (drafting a message to someone, planning work on a project), call mcp__basic-memory__read_note with people/<slug> or projects/<slug>.
If still unknown, call mcp__basic-memory__search_notes with the term as a fuzzy search.
If genuinely unknown after all of the above, ask the user, then write the answer to memory.

Example:

User: "ask todd to do the PSR for oracle"

Step 1 — read_note("claude"):
  todd → Todd Martinez ✓
  PSR  → Pipeline Status Report ✓
  oracle → not in hot cache

Step 2 — read_note("glossary"):
  oracle → Oracle Systems deal ($2.3M) ✓

Step 3 — drafting the message, need Todd's preferences:
  read_note("people/todd-martinez")

Adding memory

When the user says "remember this", "X means Y", or volunteers a fact worth keeping:

Identify the destination:
- Acronym, term, nickname, codename → glossary.
- Person → people/<slug> (create or update).
- Project → projects/<slug> (create or update).
- Decision → decisions/<YYYY-MM-DD>-<slug> (always new).
- Preference → preferences.
- Ad-hoc and dated → today's daily/<YYYY-MM-DD>.
For an update, call mcp__basic-memory__read_note to get current content. For a new note, skip this.
Build new content with YAML frontmatter and a body. Frontmatter is required for Basic Memory to index correctly.
Call mcp__basic-memory__write_note with the new content.
Promotion: if the item is used frequently or part of active work, also update the claude note to add it to the hot cache.

Daily journal

Each session of substantial work, append to today's daily note (daily/<YYYY-MM-DD>):

What happened.
What was learned.
Decisions made.

If today's note does not exist, create it. If it exists, read, append, write back.

File format

Every note must include YAML frontmatter:

---
title: Display title
type: note
permalink: short/identifier
tags: [tag, tag]
---

For daily notes, set type: daily.

Conventions

Permalinks: lowercase, hyphens. people/jane-smith, projects/project-phoenix, decisions/2026-05-03-auth-rewrite.
Capture nicknames and alternate names in glossary and in the person's file.
Capture codenames in glossary and the project's file.
Keep claude (hot cache) under ~100 lines.

What goes where

Type	Hot cache (`claude`)	Full storage
Person	Top ~30 frequent	`glossary` + `people/<slug>`
Acronym	~30 most common	`glossary` (complete)
Project	Active only	`glossary` + `projects/<slug>`
Nickname	If person is top 30	`glossary` (all)
Company context	Quick reference	`context/<topic>`
Preferences	All	`preferences`
Decisions	—	`decisions/<date>-<slug>`
Daily entries	—	`daily/<date>`

Promotion / demotion

Promote to claude when:

An item is used frequently.
It is part of active work.

Demote (remove from claude, keep in deep storage) when:

A project completes.
A person is no longer a frequent contact.
A term is rarely used.

This keeps the hot cache fresh.

Bootstrapping a new memory store

If read_note("claude") fails or returns empty, the memory store is not yet populated. Tell the user, offer to seed it, and ask:

Their name, role, current focus.
Names and roles of frequent contacts.
Active projects.
Acronyms and shorthand they use daily.

Then write to claude, glossary, preferences, and as many people/<slug> / projects/<slug> files as needed.