Local Memory Reuse

Problem

AI assistants often need to reuse user preferences, fixed rules, project context, glossary terms, task state, and history summaries across conversations. If the model has to select relevant memory from raw context every time, it spends extra token budget and may use inconsistent or noisy information.

How It Works

This skill stores memory items in a local SQLite database and organizes them by categories such as preferences, rules, project context, glossary, task state, and history summary. It returns a fixed, compact structure that can be injected directly. A typical flow is: initialize or update the database with read/write scripts in scripts/, define categories and defaults through categories/ and config/, then generate direct-injection output for the caller. The source also lists support for history references, time reminders, summary snapshots, and TTL cleanup.

Boundaries

It defaults to local, unencrypted storage, which suits offline reads, fast calls, and locally controlled workflows. If cloud sync, remote upload, or strong encryption compliance is required, an additional solution is needed. The skill does not include dynamic injection, hooks, DLLs, or drivers, and data remains only in the local SQLite database.