Memory Manager English

The Problem

Many AI Agent sessions inject long-term memory from .workbuddy/memory directly into context. As the file grows, more tokens are loaded, responses slow down, and prompt costs rise. Deleting files alone is not enough: the model needs fewer, more task-relevant memories, not more raw text. Old projects, duplicate snippets, and outdated preferences can crowd out context and steer answers away from the current task.

How It Works

Memory Manager EN treats memory as a throttle valve before context injection, organizing commands around analysis, search, compression, and backup:
- analyze, token-check, and search inspect memory size, token usage, and matching content
- load --query, rank, and throttle help load relevant memories and compress them into a budget before a session
- dedup and auto-clean default to dry-run previews, requiring explicit confirmation before changes
- Destructive operations such as import, restore, and clean require --execute to reduce accidental deletion
- export and import support memory backup and restoration
- Optional colorama, jieba, and tiktoken improve display, Chinese tokenization, and token estimation, with automatic fallback when missing

Boundaries And Notes

This skill targets .workbuddy/memory; it is not a general file manager, code debugger, or system disk cleaner. Token trends may be empty on first use, and very short memory files may be skipped during summarization. Prefer dry-run output before cleanup or restoration, and confirm the target directory is writable and memory files are UTF-8 encoded.