tutorialintermediate

The chief of staff agent Sep 2025 • Claude Agent SDK Agent Patterns Build multi-agent systems with subagents, hooks, output styles, and plan mode features.

March 8, 2026cookbook

This cookbook article introduces the Claude Agent SDK for building multi-agent systems, using a Chief of Staff agent for a startup as the primary example. It demonstrates key features including persistent memory via CLAUDE.md files, bash tool execution for Python scripts, and coordination of specialized subagents. The article progressively builds a comprehensive agent system that aggregates insights from multiple domains to provide executive summaries and actionable recommendations for a CEO managing a $10M Series A startup.

Key Points

•Use CLAUDE.md files as persistent memory and context for agents—place them in the agent's working directory and set the cwd argument in ClaudeSDKClient to enable automatic context loading
•Agents naturally prefer authoritative data sources (e.g., detailed CSVs) over high-level context; use explicit prompt instructions to guide agent behavior toward CLAUDE.md summaries when needed
•CLAUDE.md serves as an 'onboarding document' containing high-level context, company standards, and pointers to detailed data—not as hard constraints on data source selection
•Enable the Bash tool to allow agents to execute Python scripts for procedural knowledge, complex computations, data analysis, and integrations beyond native capabilities
•Organize utility scripts in accessible directories (e.g., chief_of_staff_agent/scripts) and document them in CLAUDE.md or subagent MD files for discovery and execution
•Build specialized subagents for different domains (financial analysis, recruiting, etc.) and coordinate them through a Chief of Staff agent that aggregates insights
•Use Claude Opus 4.6 for superior planning and reasoning capabilities when building complex multi-agent systems
•Implement allowed_tools configuration (e.g., Bash, Read) to control which tools agents can access and ensure security
•Design scripts with clear purposes: hiring impact modeling, candidate scoring, financial forecasting, decision matrices, and quick calculations support different agent responsibilities
•Prompt engineering matters—phrase requests to signal whether you want high-level summaries or detailed precision analysis from available data sources

Found this useful? Add it to a playbook for a step-by-step implementation guide.

Workflow Diagram

Start Process

Step A

Step B

Step C

Complete

Quality★★★★★

Concepts

Coordination Memory Systems Skills & Tools Multi-Agent Systems Agent Teams Tool Use Memory & Identity

Artifacts (4)

Chief of Staff Agent Setuppythonscript

from dotenv import load_dotenv
from utils.agent_visualizer import (
    display_agent_response,
    print_activity,
    reset_activity_context,
    visualize_conversation,
)
from claude_agent_sdk import ClaudeAgentOptions, ClaudeSDKClient

load_dotenv()

MODEL = "claude-opus-4-6"

messages = []
async with ClaudeSDKClient(
    options=ClaudeAgentOptions(
        model=MODEL,
        cwd="chief_of_staff_agent",
        setting_sources=["project"],
    )
) as agent:
    await agent.query("What's our current runway?")
    async for msg in agent.receive_response():
        print_activity(msg)
        messages.append(msg)

display_agent_response(messages)

Agent with Bash Tool Configurationpythonscript

messages = []
async with ClaudeSDKClient(
    options=ClaudeAgentOptions(
        model=MODEL,
        allowed_tools=["Bash", "Read"],
        cwd="chief_of_staff_agent",
    )
) as agent:
    await agent.query(
        "Use your simple calculation script with a total runway of 2904829 and a monthly burn of 121938."
    )
    async for msg in agent.receive_response():
        print_activity(msg)
        messages.append(msg)

display_agent_response(messages)

CLAUDE.md Template for Chief of Stafftemplate

# Chief of Staff Agent Context

## Company Overview
- **Company**: TechStart
- **Stage**: Series A ($10M raised)
- **Team Size**: 50 people
- **Current ARR**: $2.4M growing at 15% MoM

## Financial Metrics
- **Monthly Burn Rate**: $500K (gross), $235K (net)
- **Current Runway**: 20 months
- **Series B Target**: $30M

## Compensation Benchmarks
- **Senior Engineer**: $180-220K

## Available Scripts
- `scripts/hiring_impact.py`: Model hiring scenarios
- `scripts/talent_scorer.py`: Rank engineering candidates
- `scripts/simple_calculation.py`: Quick financial metrics
- `scripts/financial_forecast.py`: ARR growth projections
- `scripts/decision_matrix.py`: Strategic decision evaluation

## Data Sources
- `financial_data/burn_rate.csv`: Month-by-month burn details
- `candidate_data/`: Recruiting pipeline

## Agent Responsibilities
Coordinate specialized subagents, aggregate insights, provide executive summaries with actionable recommendations.

Utility Script Structurepythonscript

# Example: simple_calculation.py
# Performs quick financial calculations for runway, burn rate, and quarterly metrics

def calculate_runway(total_cash, monthly_burn):
    """Calculate months of runway remaining"""
    return total_cash / monthly_burn

def calculate_quarterly_burn(monthly_burn):
    """Calculate quarterly burn rate"""
    return monthly_burn * 3

def calculate_annual_burn(monthly_burn):
    """Calculate annual burn rate"""
    return monthly_burn * 12

if __name__ == "__main__":
    import sys
    if len(sys.argv) > 2:
        total_cash = float(sys.argv[1])
        monthly_burn = float(sys.argv[2])
        runway = calculate_runway(total_cash, monthly_burn)
        print(f"Runway: {runway:.1f} months")
        print(f"Quarterly Burn: ${calculate_quarterly_burn(monthly_burn):,.0f}")
        print(f"Annual Burn: ${calculate_annual_burn(monthly_burn):,.0f}")