claudekit/.claude/commands/spawn.md

/spawn

Purpose

Launch background tasks for parallel execution. Enables concurrent work on independent tasks with result aggregation.

---

Launch a background task or manage running tasks.

Spawn Operations

Launch Task

Start a new background task:

/spawn "[task description]"

Process:

1. Analyze task for parallelizability
2. Launch subagent with task
3. Return task ID for tracking
4. Continue main conversation

Output:

Spawned: Task #1
- Description: Research authentication patterns
- Status: Running
- Agent: researcher

Continue working. Use `/spawn --list` to check status.

List Tasks

Show running and completed tasks:

/spawn --list

Output:

## Active Tasks

| ID | Description | Status | Duration |
|----|-------------|--------|----------|
| #1 | Research auth patterns | Running | 2m |
| #2 | Analyze security | Complete | 5m |

## Completed Tasks (last hour)
| #2 | Analyze security | ✅ Complete | Results ready |

Collect Results

Gather results from completed tasks:

/spawn --collect

Output:

## Collected Results

### Task #1: Research auth patterns
**Status**: Complete
**Findings**:
- Pattern A: JWT with refresh tokens
- Pattern B: Session-based with Redis
- Recommendation: JWT for stateless API

### Task #2: Analyze security
**Status**: Complete
**Findings**:
- 2 high-priority issues found
- See detailed report below

Cancel Task

Stop a running task:

/spawn --cancel [id]

Task Types

Type	Best For	Agent Used
Research	Information gathering	researcher
Analysis	Code analysis	scout
Review	Code review	code-reviewer
Test	Test generation	tester
Scan	Security scanning	security-auditor

Flags

Flag	Description
--list	Show all tasks
--collect	Gather completed results
--cancel [id]	Cancel running task
--wait	Wait for all tasks to complete
--agent=[type]	Specify agent type
--priority=[high|normal]	Task priority

Usage Examples

/spawn "Research OAuth2 best practices"
/spawn "Analyze user service for performance issues"
/spawn "Review security of auth module" --agent=security-auditor
/spawn --list
/spawn --collect
/spawn --wait                    # Block until all complete

Arguments

$ARGUMENTS

If quoted text: spawn that task If flag: execute that operation

---

Parallel Workflow

Pattern: Research Phase

/spawn "Research authentication approaches"
/spawn "Analyze current auth implementation"
/spawn "Review competitor auth patterns"
# Continue other work...
/spawn --wait
/spawn --collect
# Synthesize findings

Pattern: Multi-File Review

/spawn "Review src/auth/ for security"
/spawn "Review src/api/ for performance"
/spawn "Review src/db/ for SQL injection"
/spawn --collect
# Address findings

Best Practices

1. Independent Tasks: Only spawn truly independent work
2. Clear Descriptions: Specific task descriptions get better results
3. Regular Collection: Don't let results pile up
4. Resource Awareness: Don't spawn too many concurrent tasks

Limitations

• Tasks cannot communicate with each other
• Results are collected, not streamed
• Heavy tasks may take time
• Some tasks benefit from sequential execution

Combines With

• Orchestration mode: Manages multiple spawned tasks
• /plan: Plan tasks, then spawn parallel execution
• /execute-plan: Orchestrated task execution