How it works, in three diagrams

1 · How each use case flows through the plugin

Entry point → adapter → service method → component → storage. Band width counts distinct call paths, so a use case that touches more of the system is wider. Hover a band or node for its file; click a use case to isolate it.

• Every service is built by PromptLibraryService.open(), which also runs PromptRepository.initialize() (lock + create file with header). It happens once per process, not per use case, so it isn't drawn.
• Writes re-read the CSV inside the lock, then replace the whole file atomically. That's why every write shows three storage paths.
• Bulk import is only reachable from the CLI. There is no prompt_import MCP tool.

2 · Functional decomposition of the use cases

Use case → scenario → step. Each step is one leaf, so a larger circle means more steps. Taken from commands/*.md, the SKILL.md working rules and the service code. Click to zoom in; click the background to zoom out.

3 · How the data is logically structured

Inner ring: the record, how it's stored, and the views callers get back. Outer rings: the columns, files and fields inside each. Every leaf is one element, so arc size shows how much each part contains. Click an arc to zoom; click the centre to go back.