next

Argument: $ARGUMENTS

Argument	Behavior
(none) or task	Select and implement next task (Section 2)
session	Select and implement next session (Section 3)
batch	Implement all highest-priority unblocked sessions (Section 4)
task status	Select and report next task without implementing (Section 2, stop before implementation)
session status	Select and report next session without implementing (Section 3, stop before implementation)

Always start with Section 1 (Find Tasks).

---

1. Find Tasks

ls .otto/tasks/*.json 2>/dev/null

Read each file, check for pending sessions/tasks. For each task file, also read the corresponding spec file frontmatter to get the spec name.

If no task files: "No tasks found. Run task <spec-name> to generate."

If multiple specs have pending work, use an interactive prompt:

Multiple specs have pending work:
1. {spec-name} ({spec-id}): {n} sessions pending
2. {spec-name} ({spec-id}): {n} sessions pending

Which spec should I work on?

Branch Check (Sections 2, 3, 4)

Before proceeding to any implementation section, check the current branch:

current_branch=$(git branch --show-current)

If on main or master, use an interactive prompt:

You're on '{current_branch}'. Create a feature branch or continue here?

1. Create branch: otto/{spec-id} (recommended)
2. Stay on {current_branch}

If user picks 1, create the branch and continue. If 2, proceed as-is.

---

2. Select Task

Select the next task:

1. Filter to tasks with status "pending"
2. Filter to unblocked tasks (all depends_on tasks are "done")
3. Sort by priority (lower number = higher priority)
4. Tie-break by ID (lower ID first)

If no unblocked tasks:

State	Message
All tasks "done"	"All tasks complete for {spec-name}!"
Pending but blocked	"{n} tasks blocked. Waiting on: {blocker-ids}"

Report the selected task:

Next task: {id}
Title: {title}
Priority: {priority}
Session: {session_id}

If invoked with status suffix, stop here — do NOT implement.

Implement Task

1. Update task status to "in_progress"
2. Report: "Starting task {id}: {title}"

Delegate to a subagent:

• subagent role: frontend-developer (frontend tasks) or backend-architect (backend tasks), personas defined in agents/
• Include in prompt: task context + planning workflow (see below)
• Wait for subagent to complete

Subagent prompt must include:

## Task Context
- **Task:** {task title}
- **Description:** {task description}
- **Files:** {file list}
- **Done when:** {done condition}
- **Spec path:** {spec_path from task file}

## Required Workflow

**Always plan before implementing.** Simple tasks get simple plans.

### Phase 1: Plan
1. Read the spec's Technical Design section at {spec_path}
2. Explore relevant files to understand existing patterns
3. Write your implementation plan to `.otto/plans/task-{task_id}.md`:
   - Describe your approach
   - List files to modify and key changes
   - Note any spec decisions that affect implementation

### Phase 2: Implement
4. Implement according to your plan
5. Verify the done condition is met

Keep the plan concise - it's for your reference and audit trail.

After subagent completes:

• Update task status to "done"
• Check if all tasks in session are "done"; if so, mark session "done"
• Stage: git add -u && git add .otto/
• Commit: git commit -m "otto: {spec-id} task {id} — {title}"

Report: "Task {id} complete."

---

3. Select Session

Select the next session:

1. Filter to sessions with status "pending"
2. Filter to unblocked sessions (all depends_on sessions are "done")
3. Sort by priority (lower number = higher priority)
4. Tie-break by ID (S1 before S2)

If no unblocked sessions:

State	Message
All sessions "done"	"All sessions complete for {spec-name}!"
Pending but blocked	"{n} sessions blocked. Waiting on: {blocker-ids}"

Report the selected session:

Next session: {id}
Title: {title}
Priority: {priority}
Tasks: {task_count}

If invoked with status suffix, stop here — do NOT implement.

Implement Session

1. Update session status to "in_progress"
2. Report: "Starting session {id}: {title} ({n} tasks)"

Delegate to a subagent:

• subagent role: frontend-developer (frontend tasks) or backend-architect (backend tasks), personas defined in agents/
• Include in prompt: session context + planning workflow (see below)
• Subagent implements tasks sequentially, marking each "done" as completed
• Wait for subagent to complete

Subagent prompt must include:

## Session Context
- **Session:** {session title}
- **Tasks:** {task titles with descriptions}
- **Files:** {file lists from all tasks}
- **Done when:** {done conditions for each task}
- **Spec path:** {spec_path from task file}

## Required Workflow

**Always plan before implementing.** Simple tasks get simple plans.

### Phase 1: Plan
1. Read the spec's Technical Design section at {spec_path}
2. Explore relevant files to understand existing patterns
3. Write your implementation plan to `.otto/plans/{session_id}.md`:
   - List each task you will implement
   - For each task, describe: approach, files to modify, key changes
   - Note any spec decisions that affect implementation

### Phase 2: Implement
4. Implement each task according to your plan
5. After each task: verify the done condition is met
6. Mark each task "done" as completed

Keep the plan concise - it's for your reference and audit trail.

After subagent completes:

• Update session status to "done"
• Stage: git add -u && git add .otto/
• Commit: git commit -m "otto: {spec-id} session {id} — {title}"

Report: "Session {id} complete. {n} tasks done."

---

4. Batch Mode

Select sessions to implement:

1. Filter to sessions with status "pending"
2. Filter to unblocked sessions (all depends_on sessions are "done")
3. Select only sessions at the highest priority level (lowest number)

If no unblocked sessions: Show same messages as Section 3.

If only 1 session: Continue to Section 3's "Implement Session" subsection with that session.

If multiple sessions: Launch parallel subagents.

Mark all sessions as "in_progress" before launching.

Launch all sessions in parallel as background subagents (one per session, each in an isolated/forked context):

• subagent role: frontend-developer (frontend tasks) or backend-architect (backend tasks), personas defined in agents/
• Include in prompt: session context + planning workflow (see below)
• Subagent implements tasks sequentially, marking each "done" as completed

Example (3 sessions → 3 background subagents launched together):

spawn subagent(role="frontend-developer", background=true, description="Implement S1", prompt="...")
spawn subagent(role="backend-architect", background=true, description="Implement S2", prompt="...")
spawn subagent(role="frontend-developer", background=true, description="Implement S3", prompt="...")

Subagent prompt must include:

## Session Context
- **Session:** {session title}
- **Tasks:** {task titles with descriptions}
- **Files:** {file lists from all tasks}
- **Done when:** {done conditions for each task}
- **Spec path:** {spec_path from task file}

## Required Workflow

**Always plan before implementing.** Simple tasks get simple plans.

### Phase 1: Plan
1. Read the spec's Technical Design section at {spec_path}
2. Explore relevant files to understand existing patterns
3. Write your implementation plan to `.otto/plans/{session_id}.md`:
   - List each task you will implement
   - For each task, describe: approach, files to modify, key changes
   - Note any spec decisions that affect implementation

### Phase 2: Implement
4. Implement each task according to your plan
5. After each task: verify the done condition is met
6. Mark each task "done" as completed

Keep the plan concise - it's for your reference and audit trail.

Report:

Launching {n} priority-{p} sessions in parallel:
- Session {id1}: {title1} ({n} tasks)
- Session {id2}: {title2} ({n} tasks)

Monitor and complete:

• Wait for all subagents to finish
• Mark sessions as "done"
• Stage: git add -u && git add .otto/
• Commit: git commit -m "otto: {spec-id} batch — {n} sessions at priority {p}"
• Report: "Completed {n}/{total} sessions"
• Suggest: "Run next batch again for next priority level."