winsorllc Delegate Agent
Spawn sub-agents to handle independent subtasks in parallel or sequence. This enables multi-agent workflows where specialized agents handle different aspects of a complex task.
When to Use
- Parallel execution: Multiple independent tasks can run simultaneously
- Specialized handling: Different sub-tasks need different expertise/models
- Isolation: Sub-task failures shouldn't crash the main agent
- Parallel research: Gather information from multiple sources at once
Setup
No additional setup required. This skill works out of the box.
Usage
Spawn a Sub-Agent
{baseDir}/delegate-spawn.js "research-agent" "Find information about topic" --model claude-sonnet-4-20250514
Arguments:
name: Identifier for the sub-agent (for tracking)prompt: Task description for the sub-agent--model: Optional model override (defaults to main agent's model)--timeout: Optional timeout in seconds (default: 300)
Check Sub-Agent Status
{baseDir}/delegate-status.js <job-id>
List Active Sub-Agents
{baseDir}/delegate-list.js
Kill a Sub-Agent
{baseDir}/delegate-kill.js <job-id>
Examples
Parallel Research
// Research multiple topics in parallel
const topics = ["AI agents", "Rust programming", "Docker containers"];
const agents = topics.map(topic =>
spawnSubAgent(`research-${topic}`, `Find latest developments in ${topic}`)
);
// Wait for all to complete
const results = await Promise.all(agents.map(a => a.result));
Specialized Handling
// Use different models for different tasks
await spawnSubAgent("coder", "Write a REST API", { model: "claude-sonnet-4-20250514" });
await spawnSubAgent("writer", "Write documentation", { model: "claude-haiku-20240307" });
Architecture
The delegate tool:
- Creates a new job branch
delegate/<uuid>/<agent-name> - Writes the task prompt to
logs/<uuid>/job.md - Triggers GitHub Actions to run the sub-agent
- Polls for completion and returns results
- Cleans up the temporary branch
Security Considerations
- Sub-agents inherit the same permissions as the parent
- API keys are passed to sub-agents (filtered from LLM bash output)
- Sub-agents run in isolated containers
- Activity is logged for audit purposes