prod-memory-management

What problem does it solve?

Claude and other conversational agents often misinterpret shorthand, nicknames, acronyms, and internal codenames, causing ambiguous or incorrect actions; this Skill converts those shorthand references into explicit, contextual memory so the agent can act like a knowledgeable colleague. The two-tier approach (a compact hot cache plus a deep memory directory) ensures fast decoding for common requests and unlimited scale for full organizational context.

Core Features & Use Cases

• Hot Cache (CLAUDE.md): Maintains the top ~30 people, ~30 common terms, active projects, and preferences for sub-second decoding in everyday conversations.
• Deep Memory (memory/): Stores a full glossary, individual people profiles, project pages, company context, an index, append-only logs, and weekly trend snapshots for authoritative lookups.
• Lookup Flow & Operations: Tiered lookup (CLAUDE.md → glossary → people/projects → ask user), automated ingest for bootstrapping, promotion/demotion between tiers, and weekly linting to detect contradictions and stale entries.
• Use Cases: Drafting messages with correct recipient context, assigning tasks by codename, synthesizing reports that reference historical project facts, and automating daily routines that require accurate people/project identification.

Quick Start

Ask Claude to scan your recent chat, calendar, and email to build a CLAUDE.md hot cache and populate memory/glossary.md with people, projects, and frequently used terms.