workflow.md

Workflow

The workflow has three stages: draft, read, archive.

Draft

Run paper-note-drafter on a paper PDF.

The drafter should:

1. choose a short archive name xxx;
2. create Inbox/xxx/;
3. place the PDF at Inbox/xxx/xxx.pdf;
4. extract text to Inbox/xxx/xxx.txt;
5. extract candidate figures/tables to a temporary Inbox/xxx/pdffigures2/;
6. copy only selected figures/tables into Inbox/xxx/Figure/;
7. write the first draft to both xxx-naive.md and xxx.md;
8. delete temporary extraction outputs before handoff.

The result should be:

Inbox/xxx/
  xxx.md
  xxx-naive.md
  xxx.pdf
  xxx.txt
  Figure/
    xxx-1.png

Read

Run paper-note-reader while the paper is still in Inbox/xxx/.

The reader should:

• edit xxx.md directly;
• leave xxx-naive.md unchanged;
• use xxx.txt or xxx.pdf for source verification;
• preserve figures and their Figure/xxx-N.ext references;
• compare xxx-naive.md and xxx.md during final diff review when asked.

This stage ends when xxx.md is accepted as the final note.

Archive

Run the Obsidian paper-archiver plugin.

The plugin scans Inbox/ for subfolders where the folder name and Markdown name match:

Inbox/xxx/xxx.md

For each selected paper, choose a target topic. The plugin writes:

<topic>/
  xxx.md
  PDF/
    xxx.pdf
  Figure/
    xxx-1.png

The plugin removes review artifacts:

Inbox/xxx/xxx-naive.md
Inbox/xxx/xxx.txt

It also removes unreferenced temporary files and cleans up empty Inbox directories when possible.

Boundary

Before archive, xxx-naive.md and xxx.txt are useful and should remain available.

After archive, the knowledge base should keep only:

• the final note;
• the source PDF;
• figures and tables referenced by the final note.