Memory Manager Pro

Problem

In multi-task agent workflows, context after a task is done tends to fragment across conversations, files, and ad-hoc notes. If each update requires manual checks, MEMORY.md, project indexes, type indexes, and task details quickly become unreliable. Later follow-up questions may depend on fuzzy search or full-file reads, wasting tokens and missing important files.

How It Works

Memory Manager Pro treats 完成任务并更新索引 as a unified entry point. It derives file paths from semantic task IDs such as TASK_{TYPE}_{YYYYMMDD}_{SEQ} and follows a fixed update flow: mark the task detail as completed, update the completed-task index, remove the task from the active-task index, update project and type indexes, refresh MEMORY.md and the core index, and update keyword mapping when needed. For task types such as NOVEL, CODE, DESIGN, RESEARCH, and SYSTEM, it uses the corresponding project and type index files, creating a vertical project view and a horizontal type view for cross-cutting retrieval.

Boundaries

This skill is better used as an internal task and memory indexing convention for agents, not as the executor of creative, coding, or design work. It expects callers to provide fields such as task ID, project, and result summary, and it depends on a stable directory structure. If the project structure changes frequently or a task contains large raw logs, write that content into task detail files first and then use this skill to converge the indexes.