memory-management
Two-tier memory system that makes Claude a true workplace collaborator. Decodes shorthand, acronyms, nicknames, and internal language so Claude understands requests like a colleague would. CLAUDE.md for working memory, memory/ directory for the full knowledge base.
Security Assessment
About memory-management
The memory-management skill provides a two-tier memory system designed to make Claude a highly effective workplace collaborator. It addresses the challenge of translating shorthand, acronyms, nicknames, and internal language into fully understood requests, enabling Claude to act as a knowledgeable colleague. Without this system, ambiguous references like 'ask todd to do the PSR for oracle' would be meaningless. With memory-management, Claude decodes these inputs into precise instructions, such as 'Ask Todd Martinez (Finance lead) to prepare the Pipeline Status Report for the Oracle Systems deal ($2.3M, closing Q2)', ensuring accurate understanding and action within a workplace context.
The skill features a structured memory architecture consisting of a hot cache (CLAUDE.md) and a deep memory directory (memory/). CLAUDE.md holds working memory with the top ~30 people, common acronyms, and active projects, covering approximately 90% of daily decoding needs. The memory/ directory contains the full glossary, people profiles, project details, and organizational context for comprehensive knowledge retrieval. Lookup flows prioritize the hot cache, then the full glossary, and finally prompt the user for clarification if a term is not found. This tiered approach keeps daily operations efficient while supporting unlimited scalability for long-term knowledge retention.
Use cases include automating communication, preparing accurate project summaries, and facilitating collaboration by understanding workplace-specific terminology. Target users are professionals who rely on internal shorthand and acronyms, including finance, engineering, and sales teams, who need Claude to interpret instructions correctly. This skill benefits teams seeking to maintain consistency and efficiency in internal communications, particularly where rapid understanding of internal language is critical.
FAQ
Can users directly invoke the memory-management skill?
No, this skill is not user-invocable. It operates automatically in the background to manage and decode memory for Claude.
Where is the working memory stored?
Working memory is stored in the 'CLAUDE.md' file located in the current working directory.
How does Claude handle terms not in the hot cache?
If a term is not found in CLAUDE.md, Claude searches the full glossary in memory/glossary.md. If it is still not found, Claude will ask the user for clarification and remember the new information.
What types of information are stored in deep memory?
Deep memory includes a complete glossary, detailed people profiles, project details, and organizational context, supporting full-scale knowledge retrieval beyond the hot cache.
Are there limits to the memory size?
The hot cache (CLAUDE.md) is kept lean for efficiency (~100 lines), but the deep memory directory can grow indefinitely to accommodate all workplace knowledge.
Install memory-management
Quick Setup:
- Copy the skill folder to
.claude/skills/ - Claude will automatically detect and use the skill
Repository
anthropics/knowledge-work-plugins