Each task card holds the instructions your agent will follow, along with settings that control how it branches, plans, and ships its work. New tasks land in the backlog column by default.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/cline/kanban/llms.txt
Use this file to discover all available pages before exploring further.
Two ways to create a task
Open the task form
Click the + button at the top of the backlog column, or press C from anywhere on the board. Fill in the prompt and any optional settings, then save.
Ask the sidebar agent
Open the sidebar and ask Cline (or another agent) to break a feature into tasks. Kanban injects board-management instructions into that session, so you can say things like:
“Decompose adding an auth system into tasks and add them to the board.”The agent creates the cards for you, complete with prompts and dependency links.
Task fields
| Field | Description |
|---|---|
| Prompt | The instructions sent to the agent when the task starts. |
| Base branch / ref | The git ref the worktree branches from. Defaults to the current branch. |
| Plan mode | When enabled, the agent plans before making any file edits. |
| Auto-review | Whether the agent should ship automatically when it finishes. |
Auto-review modes
Auto-review runs the shipping step without any human click. Choose the mode that matches how you want work to land.commit
commit
When the agent finishes, Kanban sends it a prompt to commit its changes directly to the base branch. Use this for small, well-scoped tasks where you trust the output.
pr
pr
When the agent finishes, Kanban sends it a prompt to push a new branch and open a pull request. Use this when you want a code review before anything merges.
move_to_trash
move_to_trash
When the agent finishes, the card moves to trash without committing or opening a PR. Use this for exploratory or research tasks where you only want to read the diff.
CLI alternative
You can create tasks from a script or another agent using thekanban task create command. Run it from inside the git repository you want to target.
| Flag | Description |
|---|---|
--prompt | Task instructions (required). |
--base-ref | Base branch or ref. Defaults to the current branch. |
--start-in-plan-mode | Start the agent in plan mode. |
--auto-review-enabled | Enable auto-review when the agent finishes. |
--auto-review-mode | One of commit, pr, or move_to_trash. |
Tips for writing good task prompts
- State the goal, not just the task. Instead of “update the login page”, write “add a “Remember me” checkbox to the login form that sets a 30-day session cookie.”
- Include acceptance criteria. Tell the agent what done looks like: “the existing test suite should still pass” or “add a unit test for the new helper.”
- Scope tasks to one concern. Smaller, focused tasks are easier to review and less likely to cause merge conflicts when running in parallel.
- Reference filenames or symbols when the agent needs to find a specific place in the code.
- Avoid open-ended instructions like “improve the codebase” — agents work best with concrete, measurable outcomes.