Goal

Goal Skill

Define goals, classify tasks by type, and execute them through superpower skills.

Task Type → Superpower Skill Mapping

When creating tasks, classify each one. The type determines which superpower skill to invoke at execution time.

Commands

Workflow

Step 1: Define

User: /goal "Add unit tests for all API endpoints"

Read state/goal.json. If one exists and active, ask about replacing.

Auto-classify each generated task:

## Goal: Add unit tests for all API endpoints

6 tasks  |  0 done  |  ○ 6 pending

P1  1. [ ] Find all untested endpoints          [explore]  brainstorming
P2  2. [ ] Write tests for GET /users            [implement]  tdd
P2  3. [ ] Write tests for POST /users           [implement]  tdd       (depends on: #2)
P2  4. [ ] Write tests for GET /items/:id        [implement]  tdd
P2  5. [ ] Write tests for PUT /items/:id        [implement]  tdd       (depends on: #4)
P3  6. [ ] Run full test suite, verify coverage  [verify]  verification

/tasks start <id> to begin, /tasks done <id> to mark complete

Each task shows its [type] and the matching superpower skill.

Step 2: Start a Task

User: /tasks start 2

## Task 2/6: Write tests for GET /users
Type: implement  |  Skill: superpowers:test-driven-development

Invoking TDD workflow:
  1. Write a failing test first (red)
  2. Write minimal code to pass (green)
  3. Refactor while tests stay green

Starting now...

Then invoke Skill("superpowers:test-driven-development") and proceed with the task.

Step 3: Task Completed

After the task is done (via its skill), user marks it:

User: /tasks done 2
Agent: ✓ Task #2 complete. 1/6 done. 5 remaining.
       Next ready: #3 (/tasks start 3), #4 (/tasks start 4)

Step 4: Goal Complete

When all tasks are completed or skipped:

1. Set goal status → completed
2. Show summary with elapsed time

Task Schema (state/goal.json)

{
  "subject": "Add unit tests for all API endpoints",
  "description": "user's full goal description",
  "status": "active",
  "created_at": "2026-07-31T06:00:00Z",
  "completed_at": null,
  "tasks": [
    {
      "id": 1,
      "subject": "Find all untested endpoints",
      "description": "Scan for endpoints without tests",
      "status": "pending",
      "priority": "high",
      "type": "explore",
      "skill": "superpowers:brainstorming",
      "depends_on": [],
      "blocks": [2, 3],
      "activeForm": "Finding untested endpoints",
      "metadata": {},
      "result": null,
      "started_at": null,
      "completed_at": null
    }
  ],
  "stats": {
    "total": 6,
    "pending": 6,
    "in_progress": 0,
    "completed": 0,
    "failed": 0,
    "skipped": 0
  }
}

Status Display

/goal status or /tasks:

## Goal: Add unit tests for all API endpoints
Status: active  |  Created: Jul 31 14:00  |  Progress: 33%

                        ┌── #2 (implement, done) [tdd]
#1 (explore, done) ─────┤
 [brainstorming]        └── #3 (implement, pending) [tdd]  blocks: #5

#4 (implement, done) [tdd]
#5 (implement, pending) [tdd] — depends on #4
#6 (verify, pending) [verification]

✓ 3 done  |  ○ 3 pending

Next: /tasks start 3 → superpowers:test-driven-development

Task Type Auto-Classification

When generating tasks from a goal description, classify each by analyzing its verb and deliverable:

If uncertain, default to implement.

Task Execution

When /tasks start <id> is invoked:

1. Read the task from state/goal.json
2. Set status → in_progress, started_at → now
3. Read the skill field
4. Invoke the superpower skill with context about the task:
   • Skill("superpowers:test-driven-development") — the skill loads and guides execution
5. The skill's workflow takes over until the task is done
6. User marks it with /tasks done <id>

If the task has type: "implement-parallel" or type: "dispatch":

• Explain that these fan out to parallel subagents
• The parent task tracks overall completion
• Sub-tasks get their own entries in the goal

Edge Cases

Key Files