memory-manager

To persist project learnings across feature implementations, you record insights as markdown files with YAML frontmatter metadata under a dedicated memory directory. The system scores these files to inject relevant context into agent prompts.

Agent prompt memory scoring is a mechanism that evaluates persisted project learnings using a defined algorithm. It ranks the files based on relevance to select the most valuable memories for injection under a project memory heading.

You format memory files using YAML frontmatter for metadata and a markdown body for the actual learnings. This structured format allows the system to parse, score, and inject the content accurately into agent prompts.

Yes, you can manage agent memory in new projects by creating a memory directory and adding an overview index file. Begin recording learnings with YAML frontmatter to bootstrap memory management and inform current decisions.

The best way to evolve agent learnings over time is to use helper functions like appendLearning and incrementUsageStat. These structured management functions update existing markdown memory files as project insights grow.

Yes, you need an index file to start managing project memory. Creating a markdown overview index in the memory directory bootstraps the system, allowing the scoring algorithm to effectively select and inject relevant learnings.