PROMPUFF

Each prompt keeps a title, description, body, a note on why it worked, a collection, tags, a favorite flag and a 1–5 usefulness rating. The Render tab lists a prompt’s {{variables}}, previews the result as values are filled in and copies it, leaving anything unfilled as a token. Search covers titles, bodies, notes and tags, #tag filters by tag, a command palette opens, copies or runs anything from the keyboard, and Quick save starts a new prompt from whatever is on the clipboard. Prompts travel as plain Markdown, one file per prompt, so import and export never need Prompuff on the other end.

Versioning is the core rule. Every distinct saved state of a prompt’s title, description, body and notes is a version with a generated note such as “Edited body (+3 −1 lines)”, a save that changes none of them creates nothing, and restoring an old version adds it as a new one, so the history never loses a step; the History tab shows each version’s line diff. Under the Avalonia app sit separate Domain, Application and Infrastructure layers, and the library is one SQLite file reached through hand-written SQL rather than an ORM. Schema migrations run against PRAGMA user_version after the database is copied to a backup, and the Markdown frontmatter reader and writer are hand-written too, so there is no YAML dependency.

It ships for Windows as a per-user Velopack installer that updates itself and for Linux as an AppImage, both self-contained. A pushed tag tests and packs both platforms, starts the AppImage under Xvfb to check that it creates its library, and opens a draft release, and headless UI tests drive the real main window through saving, rendering, copying and restarting on Windows and Linux runners. There is no account, telemetry or AI call: the only network request is the update check, and logs record prompt IDs, never their text.