/start

/task:start $ARGUMENTS

Start work on ClickUp task with superpowers-driven planning.

Workflow

1. Parse task ID

Parse from argument:

• CU-xxx → use directly
• https://app.clickup.com/t/xxx → extract ID
• #xxx or just xxx → prepend CU-

If no argument provided, ask user for task ID.

2. Fetch task via clickup-task-agent

Use Task tool with subagent_type="task:clickup-task-agent":

"Fetch ClickUp task CU-{id}:
1. Get task details (name, description, acceptance criteria)
2. Capture current status (for potential rollback)
3. Update status to 'in progress'
4. Assign to me

Return: task summary, suggested branch name, original_status"

Store original_status in conversation context for rollback on abort.

3. Enter Plan Mode + Brainstorming

First, use EnterPlanMode tool to enter plan mode (creates plan file at .claude/plans/<name>.md).

If already in plan mode: Skip EnterPlanMode, use existing plan file.

Then invoke brainstorming (which writes to the active plan file):

Skill tool with skill="superpowers:brainstorming", args:

"Implement ClickUp task CU-{id}: {task-name}

## ClickUp Context
{description from subagent}

## Acceptance Criteria
{criteria from subagent}

## Branch Convention
Use branch: feat/CU-{id}-{slug} (or fix/ for bugs)"

Brainstorming will explore context, ask questions, and output design doc to the plan file.

4. Review Loop

Track iteration_count = 0.

REPEAT (max 5 iterations):

1. Increment iteration_count
2. Invoke Skill("kas:review-plan")
3. If APPROVED: call ExitPlanMode, proceed to step 5
4. If iteration_count >= 5: force user to choose override or abort (no more revisions)
5. If NEEDS REVISION or BLOCKED:
   • Present findings to user (quote critical/high issues verbatim)
   • State overall verdict
   • Use AskUserQuestion tool with options:
     • Revise - refine plan with feedback
     • Override - proceed despite issues (user accepts risk)
     • Abort - stop workflow
   • If Revise: re-invoke brainstorming with feedback context, continue loop
   • If Override: call ExitPlanMode, proceed to step 5
   • If Abort: revert ClickUp status to original_status, stop workflow

On abort: Use clickup-task-agent: "Update task CU-{id} status to {original_status}"

On ExitPlanMode: User sees plan + review findings together and confirms readiness.

5. Implementation Choice

After plan approval, present options:

Plan approved. Choose implementation approach:
1. Subagent-Driven (this session) - Fresh subagent per task
2. Parallel Session - Create worktree, implement in new session

Option 1: Invoke Skill("superpowers:subagent-driven-development")

Option 2:

1. Invoke Skill("superpowers:using-git-worktrees")
2. Save session context to .claude/task-session.json:

   {
     "taskId": "CU-xxx",
     "branch": "feat/CU-xxx-slug",
     "worktree": ".worktrees/CU-xxx-slug/",
     "planFile": ".claude/plans/xxx.md",
     "status": "awaiting-implementation"
   }
   

3. Output cd command and next session prompt
4. STOP (user runs /clear, uses executing-plans in new session)

Error Handling

Error	Action
No argument	Ask for task ID
Task fetch fails	Show error, suggest retry
Task not found	Show error with valid URL format
EnterPlanMode fails	Show error, suggest manual plan creation
Already in plan mode	Use existing plan file, continue workflow
Brainstorming unavailable	Fall back to /kas:start with ClickUp context
Brainstorming crashes	Save partial plan, offer retry or manual edit
Review agent fails	Ask user: proceed without review (warn) or retry
User aborts	Revert ClickUp status, stop workflow
Max iterations (5)	Force choice: override or abort

Notes

• Worktree creation happens AFTER plan approval
• Task ID stored in session context for later commands
• Plan file path provided by plan mode context