ワンクリックで
next
Pick or implement the next task or session. Use when continuing work, picking the next task, or implementing tasks from a task list.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Pick or implement the next task or session. Use when continuing work, picking the next task, or implementing tasks from a task list.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Pick or implement the next task or session. Use when continuing work, picking the next task, or implementing tasks from a task list.
Reviews code changes for bugs with P0-P2 prioritized feedback. Uses parallel subagents for thorough analysis, then creates fix plans. Use when reviewing code, finding bugs, checking quality, or before merging. Use review fix to implement fixes.
Writes product specifications through collaborative interview with web research. Use when planning, gathering requirements, designing new features, or creating a spec/PRD.
Generates parallelizable task lists from specs. Breaks specs into atomic, prioritized tasks with dependencies. Use when you have a spec and need tasks, implementation breakdown, or a work plan.
Runs lint, type check, tests, and visual verification. Auto-detects tools. Use when running tests, linting, type checking, or writing tests.
Synthesizes code docs into user-facing HTML summary. Creates semantic narrative explaining what changed and why it matters. Use when creating PR summaries, release notes, or change overviews.
| name | next |
| description | Pick or implement the next task or session. Use when continuing work, picking the next task, or implementing tasks from a task list. |
| argument-hint | ["task | session | batch"] |
Argument: $ARGUMENTS
| Argument | Behavior |
|---|---|
(none) or task | Select and implement next task (Section 2) |
session | Select and implement next session (Section 3) |
batch | Implement all highest-priority unblocked sessions (Section 4) |
task status | Select and report next task without implementing (Section 2, stop before implementation) |
session status | Select and report next session without implementing (Section 3, stop before implementation) |
Always start with Section 1 (Find Tasks).
ls .otto/tasks/*.json 2>/dev/null
Read each file, check for pending sessions/tasks. For each task file, also read the corresponding spec file frontmatter to get the spec name.
If no task files: "No tasks found. Run task <spec-name> to generate."
If multiple specs have pending work, use an interactive prompt:
Multiple specs have pending work:
1. {spec-name} ({spec-id}): {n} sessions pending
2. {spec-name} ({spec-id}): {n} sessions pending
Which spec should I work on?
Before proceeding to any implementation section, check the current branch:
current_branch=$(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.
Select the next task:
depends_on tasks are "done")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.
Delegate to a subagent:
frontend-developer (frontend tasks) or backend-architect (backend tasks), personas defined in agents/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:
git add -u && git add .otto/git commit -m "otto: {spec-id} task {id} — {title}"Report: "Task {id} complete."
Select the next session:
depends_on sessions are "done")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.
Delegate to a subagent:
frontend-developer (frontend tasks) or backend-architect (backend tasks), personas defined in agents/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:
git add -u && git add .otto/git commit -m "otto: {spec-id} session {id} — {title}"Report: "Session {id} complete. {n} tasks done."
Select sessions to implement:
depends_on sessions are "done")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):
frontend-developer (frontend tasks) or backend-architect (backend tasks), personas defined in agents/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:
git add -u && git add .otto/git commit -m "otto: {spec-id} batch — {n} sessions at priority {p}"next batch again for next priority level."