ganban
git-based Kanban TUI
What is this?
Note: this is the plan. half of this isn’t working yet. half of it is, and I’m not saying which half.
ganban is a Kanban board that lives in git. It sits in an orphan branch so web
UIs don’t offer to merge it. In ganban, our tasks stay with the code and we
don’t need to rely on external services and API keys. It’s just files in a dir.
The board’s columns are directories named id.slug like 1.backlog. They
don’t need a numeric id, they’re ordered alphanumerically so the TUI has the
same order as ls -l. The TUI will renumber them for you if you move columns
around.
Cards on a board are symlinks to Markdown docs in the ./.all dir, which holds
the actual data. The links are ordered in the same way as cards. Keeping them in
.all gives them stable IDs, so you can link to them. ganban will round up
stray files and replace them with a link, so you don’t need to remember ln’s
weird syntax.
The first # heading, if it’s present, will override the title of a card. Any
## subheading becomes a section, and the TUI can choose a custom editor for
sections with special names. This is how comments and todo lists work.
You can also add metadata like labels, a due date, assignee, story points - or
anything else you like - using front-matter. Metadata works on columns too, by
editing the index.md file in their dir. And the board of course, which can
have a title and settings in its metadata.
When running as a service, ganban will sync from remotes that the repo has,
periodically pushing to upstream and resolving conflicts the best it can. This
creates a kind of mesh network for your board.
If you don’t like the terminal, you can use the web UI (which looks just like
the TUI.) If you’re a robot that isn’t good at UIs, ganban -d will sync in the
background as you work. If the daemons haunt you, ganban --sync for manual
syncing. If you’re a bot, there’s context-friendly instructions for use in the
board’s default index.md.
Overview
Open a repo with ganban and it reads from a ganban orphan branch. The board
is just directories (columns) containing symlinks (cards) pointing to
canonical card files. Everything is plain markdown, works with any text
editor, and syncs via git.
Data Model
Cards
Canonical card files live in .all/ with 3-digit numeric IDs:
.all/
001.md
002.md
003.md
Card format:
# Title of the card
Main description goes here, displayed in the body.
## Notes
Subsequent sections become sidebar panels in the TUI.
## Comments
**@user 2024-01-15**: Append-only, TUI adds name + date automatically.
## Links
- blocks [#42](../.all/042.md)
- see also [#7](../.all/007.md)
- Title from first
# heading, not front-matter - Content before first
##= main body - Each
##= collapsible sidebar section ## Comments= special append-only widget## Links= relationships, rendered specially- Cross-references: type
#42, stored as[#42](../.all/042.md)so links work everywhere - Front-matter only for optional extras (tags, color, custom fields)
Tags on separate lines for clean merges:
tags:
- first
- second
Columns
Directories with 1-digit numeric prefix for ordering:
1.backlog/
2.in-progress/
3.done/
- Title from
index.mdfirst# heading, or normalized from dirname in-progress→In progress- Hidden columns start with
.(e.g.,.all) - Optional
index.mdfor description and metadata
Board Layout
Cards appear in columns as symlinks with 2-digit position prefix:
1.backlog/
01.fix-login-bug.md -> ../.all/001.md
02.add-feature.md -> ../.all/003.md
2.in-progress/
01.refactor-api.md -> ../.all/002.md
- Symlink name = position + slug (derived from card title)
- Moving card between columns = delete + create symlink
- Reordering within column = rename symlinks
- Archiving = just delete the symlink, card stays in
.all/ - Orphaned cards (in
.all/but not linked) accessible via TUI
Numbering
Default digits: 3 for .all/, 1 for columns, 2 for symlinks. Can extend with
hex or base64 if needed, but integers are the default.
Git Integration
Reading/Writing Without Checkout
Operates on the ganban branch without checking it out, using git plumbing:
git show ganban:pathto readgit hash-object,git mktree,git commit-treeto write- Or GitPython equivalents
Sync Workflow
- Commit before pull (TUI enforces this)
git pull --rebaseon tasks branch- Reload board
- If conflicts, show conflict UI
Conflict Handling
- Content conflicts: git markers show in markdown, user edits them out
- Broken symlinks: highlight red, user picks destination or deletes
- Column renames: broken links surface as red cards, same fix
Multi-Remote Sync
- One upstream (push target)
- Optional peer remotes (extra fetch sources)
- Fetch all remotes periodically
- On detecting changes: commit local, rebase, push upstream
- Changes propagate opportunistically through whoever can reach upstream
Philosophy
- Permissive: don’t be prescriptive, let users mess with files
- Resilient: handle broken states gracefully, offer to fix
- Minimal: derive what you can (titles, order), front-matter is opt-in
- Portable: works with any markdown renderer, any git host, any editor