cline-messages
v0.2.0
Published
Fork of cline/kanban adding MCP tools (kanban_set_task_output / kanban_read_linked_task_outputs) for passing data between linked tasks.
Maintainers
Readme
cline-messages
This is a fork of cline/kanban. The vast majority of the code, design, and credit belongs to the Cline team. Their upstream repository is the authoritative source — please star and follow it.
This fork adds two MCP tools,
kanban_set_task_outputandkanban_read_linked_task_outputs, so an agent running inside one task card can hand a string output to an agent running in a linked downstream task card. Seesrc/cline-sdk/kanban-builtin-mcp-server.tsandsrc/mcp-server/kanban-mcp-stdio.tsfor the added code. The fork is licensed under Apache-2.0 (seeLICENSEandNOTICE).Install this fork with
npm install -g cline-messages. It installs thecline-messagesbinary; your existing upstreamkanbaninstall is untouched.
npx kanban (Research Preview)
A replacement for your IDE better suited for running many agents in parallel and reviewing diffs. Each task card gets its own terminal and worktree, all handled for you automatically. Enable auto-commit and link cards together to create dependency chains that complete large amounts of work autonomously.
[!WARNING] Kanban is a research preview and uses experimental features of CLI agents like bypassing permissions and runtime hooks for more autonomy. We'd love your feedback in #kanban on our discord.
1. Open kanban
# Run directly (no install required)
npx kanban
# Or install globally
npm i -g kanban
kanbanRun this from the root of any git repo. Kanban will detect your installed CLI agent and launch a local running webserver in your browser. No account or setup required, it works right out of the box.
2. Create tasks
Create a task card manually, or open the sidebar chat and ask your agent to break work down into tasks for you. Kanban injects board-management instructions into that session so you can simply ask it to add tasks, link tasks, or start work on your board.
3. Link and automate
⌘ + click a card to link it to another task. When a card is completed and moved to trash, linked tasks auto-start. Combine with auto-commit for fully autonomous dependency chains: one task completes → commits → kicks off the next → repeat. It’s a pretty magical experience asking your agent to decompose a big task into subtasks that auto-commit - he’ll cleverly do it in a way that parallelizes for maximum efficiency and links tasks together for end-to-end autonomy.
Trigger mode: start the next task on Review instead of Trash
By default, a linked downstream task only auto-starts after the upstream task is moved from Review → Trash (blue arrow, ⌘ + click). If you'd rather have the next task kick off as soon as the upstream lands in the Review column — useful when the upstream uses autoReviewMode: commit or pr and you want to keep its work around — hold ⌘ + Shift while clicking to draw the link. The arrow renders in purple to indicate Review-mode triggering. Trash and Review modes are per-link, so different chains on the same board can use different rules.
4. Start tasks
Hit the play button on a card. Kanban creates an ephemeral worktree just for that task so agents work in parallel without merge conflicts. Under the hood, it also symlinks gitignored files like node_modules so you don't have to worry about slow npm installs for each copy of your project.
[!NOTE] Symlinks (symbolic links) are special "shortcuts" pointing to another file or directory, allowing access to the target from a new location without duplicating data. They work great in this case since you typically don't modify gitignored files in day-to-day work, but for when you do then don't use Kanban.
As agents work, Kanban uses hooks to display the latest message or tool call on each card, so you can monitor hundreds of agents at a glance without opening each one.
5. Review changes
Click a card to view the agent's TUI and a diff of all the changes in that worktree. Kanban includes its own checkpointing system so you can also see a diff from the last messages you've sent. Click on lines to leave comments and send them back to the agent.
To easily test and debug your app, create a Script Shortcut in settings. Use a command like npm run dev so that all you have to do is hit a play button in the navbar instead of remembering commands or asking your agent to do it.
6. Ship it
When the work looks good, hit Commit or Open PR. Kanban sends a dynamic prompt to the agent to convert the worktree into a commit on your base ref or a new PR branch, and work through any merge conflicts intelligently. Or skip review by enabling auto-commit / auto-PR and the agent ships as soon as it's done. Move the card to trash to clean up the worktree (you can always resume later since Kanban tracks the resume ID).
7. Keep track with git interface
Click the branch name in the navbar to open a full git interface to browse commit history, switch branches, fetch, pull, push, and visualize your git all without leaving Kanban. Keep track of everything your agents are doing across branches as work is completed.
