Files
goodbuddy/docs/features/magic-notes
mesalogo a3b7a27045 fix: preserve drafts, canvas edits, and remote Continue tools
Keep paragraph formatting and checklist todos when removing canvas pages,
and preserve accepted undo/redo history after oversized input is rejected.
Merge restored queued text into the latest conversation draft and retain
image delivery when OCR returns image references and generated warnings only.

Deliver session MCP servers to bridged remote Continue Execute while keeping
Ask read-only and preserving local model-profile isolation. Add regression
coverage and update feature documentation. Correct stale conversation-limit
and forced-color assertions and the measured Electron layout test deadline.

Validation: 4,826 tests passed, 67 skipped; full typecheck and lint passed.
Current-source Linux x64 Execute/Ask/Execute MCP delivery passed, using 10
real text-provider requests and no image-provider requests. Image generation
itself was not exercised. Staged whitespace checks passed before commit.

Release note: 修复恢复队列时覆盖新输入、画布删页丢失清单格式及超限输入清空撤销历史的问题;图片 OCR 无正文时保留原图发送方式,远程 Continue Execute 正确保留会话 MCP 工具。
2026-09-19 23:09:48 +08:00
..

Magic Notes

Magic Notes is the local notes, checklist and AI-comment workbench. Desktop actions and authorized Agent/MCP tools use the same desktop store: SQLite for metadata and indexes, and local files for entry bodies and binary assets. The open workbench subscribes to successful writes so users can see external changes without leaving the page or clicking a refresh button.

The workbench opens on a responsive note-card overview. Opening a note replaces the overview with a continuous record stream, a narrow left index and independent right AI comments. Both side panes can collapse. Index items show one real first-page canvas thumbnail with page count, or a text summary; clicking scrolls without replacing the editor or losing a draft. Opening a note shows a blank composer above the history, with text/canvas selection and Save. Creating a record resets that composer; editing a saved record requires its explicit Edit action. The separate To-dos view keeps search, status filters and grouped tasks on one page, with direct completion and a stable left list/right detail layout for instructions, source content and AI comments. Narrow containers use on-demand detail with an explicit return to the mounted list. One content-sized button switches to the other view in the page header immediately before New note; status filters sit beside search and wrap when needed. Note detail removes the surrounding frame and stream padding while keeping editor borders and the AI divider. Returning from a source note restores the task context and protects unsaved editor changes. A successful save followed by no further edits does not prompt again on return. Narrow note layouts use an index drawer and bounded bottom AI pane.

Documents

  • Technical design: file storage, schema 38 migration, backup boundaries, database notifications, process boundaries, canvas and AI contracts, refresh lifecycle and regression coverage.
  • Progress: implementation and validation evidence.
  • UI design system: authoritative workbench interaction rules, including selection and draft preservation.
  • Canvas integration: React controller API, PeopleLib core limits, asset adaptation and rendering lifecycle.

An entry is a saved rich-text or paged-canvas record within a note. Its body lives in notes/<noteId>/entries/<entryId>.json beside note-local binary assets; note.json is a derived membership manifest. SQLite retains metadata, indexes, revisions, todos and comments. API reads hydrate assets back into the existing content format. A todo is a checklist item derived from rich text or the canvas Quill flow. An unsaved draft belongs to the open editor; it is not written to storage by a background refresh.

Backups must coordinate SQLite and the adjacent notes/ tree while note writes are paused or the app is closed. A database-only backup cannot restore these bodies. There is no product one-click backup action; see the backup contract.

Canvas Scope

The integrated PeopleLib editor uses Fabric for paged annotations and Quill for body text flowing across pages. It supports paper templates, pen/highlighter, object erasing and selection/transforms, floating text, images, PDF backgrounds with native text extraction, and raster PDF export. Saves are manual. Undo/redo belongs to the current annotation page or Quill body mode; it is not a global document history. There is no infinite canvas or PNG download button. Editors and saved viewers support view zoom and responsive Fit width without changing saved content or export dimensions; see the canvas zoom reference.

Drafts, saved entries and todos sourced from canvases support AI comments using the default model's image-input capability. Image-capable models receive complete page captures and extracted text. Text-only models receive an explicitly labelled text fallback; a purely visual canvas entry or draft reports that an image-capable default model is required. Canvas comments are requested manually in immediate mode, or after a manual save in after-save-auto mode. Analysis failure does not undo a successful save. Detailed retention rules are in the AI contract.

Canvas implementation is complete and final validation is in progress. The progress record distinguishes confirmed production-path evidence from the pending final repository checks.