---

name: claude-agent description: > Spawn and manage Claude sub-agents for parallel or delegated tasks. WHEN: User needs parallel operations, multi-step autonomous research, or wants to delegate complex tasks to sub-agents. Use when task can be parallelized or requires independent exploration. WHEN NOT: Simple single-step operations, sequential dependencies, specific file reads (use fs_read_file), writing code directly (do it yourself). version: 0.1.0

Claude Agent - Sub-Agent Delegation

Core Concept

mcp__plugin_kg_kodegen__claude_agent spawns independent Claude sub-sessions that can execute tasks autonomously. Each agent has its own conversation context, can use tools, and returns a final report. Perfect for parallel research, independent code analysis, or complex multi-step delegations.

Five Actions

SPAWN (Default)

Create a new agent session with initial prompt.

SEND

Send additional prompt to existing agent.

READ

Read current output from agent.

LIST

List all active agent sessions.

KILL

Terminate agent session and cleanup.

Key Parameters

Usage Examples

Spawn Research Agent

Copy{
  "action": "SPAWN",
  "prompt": "Research all error handling patterns in this codebase. Return a summary of patterns found with file locations.",
  "max_turns": 15
}

Parallel Agents for Different Tasks

Copy// Agent 0: Research
{
  "agent": 0,
  "prompt": "Find all API endpoints and document their signatures"
}

// Agent 1: Analysis (concurrent)
{
  "agent": 1,
  "prompt": "Analyze test coverage and identify untested code paths"
}

Restricted Agent (Read-Only)

Copy{
  "prompt": "Review this codebase for security vulnerabilities",
  "allowed_tools": ["fs_read_file", "fs_search", "fs_list_directory"],
  "disallowed_tools": ["terminal", "fs_write_file", "fs_delete_file"]
}

Background Agent with Timeout

Copy{
  "prompt": "Deep dive into the authentication system architecture",
  "await_completion_ms": 60000,
  "max_turns": 20
}

Check Agent Progress

Copy{"action": "READ", "agent": 0}

List All Agents

Copy{"action": "LIST"}

Terminate Agent

Copy{"action": "KILL", "agent": 0}

When to Use What

Best Practices

1. Be specific in prompts - Tell agent exactly what to return
2. Specify output format - Request structured results
3. Use tool restrictions - Limit agent capabilities when appropriate
4. Launch concurrently - Multiple agents in single message for parallelism
5. Trust agent output - Results are generally reliable

Remember

• Agents are stateless - each invocation is independent
• Agent results are not visible to user - you must summarize
• Prompts should be highly detailed - agent works autonomously
• Launch multiple agents concurrently for parallel work
• Specify if agent should research only vs write code