This week was about making Seed feel safer and more practical in day-to-day publishing: fewer dead ends in the editor, clearer desktop navigation, and a stronger release pipeline for Windows builds.

What I shipped

Simplified document sharing and cleanup on desktop: the Share menu became “Copy link”, deleted cards are removed more reliably, and embed card clicks route through the same document-card navigation path. See PR #797, commit 1f8fd7b7, and commit 7d84ca52.

Polished desktop chrome: video controls now apply to actual players, the titlebar aligns with the locked sidebar width, bookmark labels truncate cleanly, and tabs can show a people count. See PR #804, PR #805, PR #806, and commit 6affde90.

Added DigiCert KeyLocker signing for Windows builds, then tuned the development workflow around it. See PR #811 and commit 041cb6b0.

Fixed editor rough edges around image captions, query placeholders, and autosave. Caption clicks no longer select the whole image block, query new-doc placeholders hide when a banner is present, and draft snapshots preserve recoverable edits during autosave writes. See PR #815, PR #816, and PR #818.

Replaced separate move and republish flows with a unified destination dialog on desktop. This gave document relocation a clearer mental model and added tests for child path handling and republish refs. See PR #817.

Added writer promotion for site members, including account canonicalization so delegate aliases dedupe to root accounts and writer grants win over regular membership rows. See PR #821.

Added frontend UI polish agent guidance and pinned formatting behavior across the repo. See commit 297dcb5b and commit ad831d68.

Notes

A lot of the work was small on its own, but it all pointed at the same thing: publishing should feel continuous. The user should be able to copy a link, move a document, edit a caption, or promote a collaborator without wondering which subsystem they accidentally crossed into.

My question from this week:

How many separate document actions can we collapse into one destination model before it becomes too abstract?