Task

Generate Mode

Usage: /task <spec-name-or-id>

Resolve Spec Argument

1. Try reading .otto/specs/$ARGUMENTS.md (exact ID match)
2. If not found, list .otto/specs/*.md and read frontmatter of each:
   • Match name field (case-insensitive substring)
3. If exactly one match: use that spec
4. If multiple matches: use AskUserQuestion to disambiguate, showing name and ID for each
5. If no matches: "No spec matching '{argument}'. Run /spec list to see available specs."

Task Design Principles

Each task should be an atomic unit of work:

• One deliverable: Clear output that indicates completion
• Verifiable: Specific "done" condition
• Atomic scope: Completable in one agent session
• File limit: Ideally ≤3 files modified (split larger tasks)
• Minimal dependencies: Only add depends_on when task truly cannot start without another

Workflow

1. Read Spec

Resolve the spec argument (see Resolve Spec Argument above), then read the spec file.

If not found after resolution, list available specs:

ls .otto/specs/*.md 2>/dev/null

If spec status is not "approved", report: "Spec has status '{status}'. Only approved specs can generate tasks."

2. Analyze & Break Down

Break spec into discrete, implementable tasks following the design principles above.

Assign priorities:

• 0 = Critical (blocks other work)
• 1 = High (core functionality)
• 2 = Normal (default)
• 3 = Low (can defer)
• 4 = Nice to have

Assign task type:

• frontend - UI components, styling, client-side state, React/Vue/etc., browser APIs
• backend - APIs, database, server logic, authentication, infrastructure

2b. Group into Sessions

Group related tasks into sessions — units of work that can be completed in one agent session.

Grouping criteria:

• Component cohesion: Tasks affecting the same module/directory
• File overlap: Tasks modifying shared files
• Sequential dependencies: Dependent tasks that naturally chain

Session sizing: 2-5 tasks per session (minimum 1)

Session dependencies:

• Session A depends on Session B if any task in A depends on any task in B
• Session priority = minimum priority of its tasks

3. Save Draft

Write to .otto/tasks/{spec-id}.json:

{
  "spec_id": "{spec-id}",
  "spec_path": ".otto/specs/{spec-id}.md",
  "sessions": [
    {
      "id": "S1",
      "title": "Session title",
      "status": "pending",
      "priority": 0,
      "depends_on": [],
      "task_ids": ["1", "2"]
    }
  ],
  "tasks": [
    {
      "id": "1",
      "title": "Task title",
      "description": "Success: [done condition]. Files: [paths]. Scope: [estimate]",
      "status": "pending",
      "priority": 1,
      "type": "frontend | backend",
      "depends_on": [],
      "session_id": "S1"
    }
  ]
}

4. Review Task List

Launch principal-engineer subagent with Task tool (forked context):

• Prompt includes: task list JSON, full spec content, review criteria below
• Subagent has no access to conversation - all context in prompt

Review Criteria:

• Atomicity: Tasks too large (should split)
• Ordering: Tasks in wrong order, should be resequenced
• Dependencies: Missing dependencies that will cause blocked work
• Circular dependencies: Will cause deadlock
• Completeness: Tasks missing from spec
• Verifiability: Missing "done when" conditions

Finding format (P0 = critical, P1 = important, P2 = minor):

### [P{0-2}] {title}
**Tasks:** {task IDs affected}
**Issue:** {what's wrong}
**Suggestion:** {split, merge, reorder, add dependency}
**Alternatives:** {if non-obvious, options}

Wait for review to complete.

5. Interview User on Findings

If no findings, skip to step 6.

For each finding (highest priority first):

1. Present the finding with its priority level
2. If suggestion is clear: AskUserQuestion with "Accept", "Reject", "Modify"
3. If alternatives exist: AskUserQuestion with the options
4. If accepted: Update the task list JSON file with the change
5. If rejected: Skip to next finding
6. If modify: Apply user's modified version to the JSON file

After processing all findings, continue to step 6.

6. Approval

Show sessions with nested task tables:

Proposed sessions for {spec-name}:

## Session S1: {session-title} (P{priority})
| ID | Title | Type | Priority | Depends On |
|----|-------|------|----------|------------|
| 1 | Setup project | backend | P0 | - |
| 2 | Core types | backend | P0 | 1 |

## Session S2: {session-title} (P{priority}, depends on S1)
| ID | Title | Type | Priority | Depends On |
|----|-------|------|----------|------------|
| 3 | Core feature | frontend | P1 | 2 |
| 4 | Feature tests | frontend | P1 | 3 |

Note parallelism:

• "Parallel sessions: S{ids} can run concurrently"
• "Sequential sessions: S1 → S2 → S3"

Use AskUserQuestion with options:

• "Approve"
• "Request changes"

Revise until approved.

Report: "Created {n} tasks in {m} sessions for {spec-name}"

7. Next Steps

Offer to start:

"Run /next to implement the first session, or /next batch to run all unblocked sessions in parallel."