Agent Builder Skill

Metadata

• Name: agent-builder
• Trigger: Auto-triggered when building AI agents with DAPR
• Category: AI Agent Development

Description

Provides intelligent assistance for building AI agents with the DAPR Agents framework. Validates configuration, suggests patterns, and ensures best practices.

Auto-Trigger Patterns

This skill activates when detecting:

• Creation of agent files (*_agent.py, *agent*.py)
• Import of dapr_agents or dapr-agents
• Agent configuration files
• Tool definitions with @tool decorator
• Workflow-backed agent patterns

Capabilities

1. Agent Configuration Validation

Validates agent setup:

• Required imports present
• Agent name and role defined
• Instructions are comprehensive
• Model configuration is valid
• Tools are properly decorated

2. Pattern Recommendations

Suggests appropriate patterns based on use case:

• AssistantAgent: Basic interactive agents
• DurableAgent: Fault-tolerant long-running tasks
• AgentService: Headless REST API agents
• Multi-Agent: Complex orchestration

3. Tool Integration Check

Validates tool definitions:

• @tool decorator present
• Docstrings for LLM understanding
• Pydantic input validation
• Async for I/O operations
• Error handling

4. Memory Strategy Guidance

Recommends memory configuration:

• Short-term (conversation context)
• Long-term (DAPR state store)
• Vector memory (RAG scenarios)

5. LLM Configuration Check

Validates LLM setup:

• API key environment variables
• Model name configuration
• Temperature settings
• Token limits

Validation Rules

Agent Definition

# Required elements:
from dapr_agents import AssistantAgent

agent = AssistantAgent(
    name="...",           # Required: unique identifier
    role="...",           # Required: agent's role
    instructions="...",   # Required: system prompt
    tools=[...],          # Optional: list of tools
    model="...",          # Optional: LLM model (default: gpt-4o)
)

Tool Definition

# Recommended pattern:
from dapr_agents import tool
from pydantic import BaseModel, Field

class ToolInput(BaseModel):
    """Input validation with descriptions."""
    param: str = Field(..., description="Parameter description")

@tool
async def my_tool(input: ToolInput) -> str:
    """
    Tool description for LLM.

    Args:
        input: Validated input parameters

    Returns:
        Tool result
    """
    # Implementation
    return result

Workflow Agent

# Durable agent pattern:
from dapr.ext.workflow import workflow, activity

@activity
async def agent_activity(ctx, data: dict) -> dict:
    """Durable activity with agent logic."""
    pass

@workflow
def agent_workflow(ctx: DaprWorkflowContext, data: dict):
    """Workflow orchestrating agent activities."""
    result = yield ctx.call_activity(agent_activity, input=data)
    return result

Checklist

When building agents, ensure:

Agent Setup

• Agent has unique name
• Clear role describing purpose
• Detailed instructions for LLM
• Appropriate model selected
• temperature set for use case

Tools

• All tools have @tool decorator
• Docstrings explain tool purpose
• Complex inputs use Pydantic models
• Async used for I/O operations
• Errors handled gracefully

Memory

• Short-term memory for context
• Long-term memory via DAPR state
• Vector memory if using RAG

Security

• API keys in environment variables
• Input validation on tools
• No sensitive data in logs
• Rate limiting considered

Durability (for DurableAgent)

• Activities are idempotent
• Workflow handles failures
• State is checkpointed
• Retry policies defined

Common Issues

1. Missing Tool Docstrings

# Bad: LLM won't know when to use
@tool
def my_tool(x: str) -> str:
    return x.upper()

# Good: Clear purpose
@tool
def my_tool(x: str) -> str:
    """Convert text to uppercase for formatting."""
    return x.upper()

2. Sync Tools with I/O

# Bad: Blocks event loop
@tool
def fetch_data(url: str) -> str:
    return requests.get(url).text

# Good: Async for I/O
@tool
async def fetch_data(url: str) -> str:
    async with httpx.AsyncClient() as client:
        return (await client.get(url)).text

3. Hardcoded API Keys

# Bad: Security risk
client = OpenAI(api_key="sk-...")

# Good: Environment variable
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))

4. Missing Input Validation

# Bad: No validation
@tool
def process(data: dict) -> str:
    return data["key"]  # May fail

# Good: Pydantic validation
class ProcessInput(BaseModel):
    key: str = Field(..., min_length=1)

@tool
def process(input: ProcessInput) -> str:
    return input.key

Agent Patterns Reference

Interactive Assistant

agent = AssistantAgent(
    name="assistant",
    role="Helpful Assistant",
    instructions="Help users with their questions.",
    tools=[tool1, tool2],
    model="gpt-4o"
)

Durable Research Agent

@activity
async def research(ctx, topic: str) -> str:
    agent = AssistantAgent(...)
    return await agent.run(f"Research: {topic}")

@workflow
def research_workflow(ctx: DaprWorkflowContext, topic: str):
    result = yield ctx.call_activity(research, input=topic)
    return result

REST API Agent

@app.post("/chat")
async def chat(request: ChatRequest):
    response = await agent.run(request.message)
    return {"response": response}

Multi-Agent System

specialists = {
    "technical": technical_agent,
    "support": support_agent,
}

router = AssistantAgent(
    name="router",
    instructions="Route to appropriate specialist..."
)

Environment Variables

Required for agent operation:

# LLM Configuration
OPENAI_API_KEY=sk-...           # OpenAI API key
AZURE_OPENAI_API_KEY=...        # Azure OpenAI key
AZURE_OPENAI_ENDPOINT=...       # Azure endpoint
LLM_MODEL=gpt-4o                # Default model

# DAPR Configuration
DAPR_HTTP_PORT=3500             # DAPR sidecar port
DAPR_GRPC_PORT=50001            # DAPR gRPC port

# State Store
STATE_STORE_NAME=statestore     # DAPR state store name

# Pub/Sub
PUBSUB_NAME=pubsub              # DAPR pub/sub name

Templates

Use the /dapr:agent command to generate agent templates:

/dapr:agent assistant my-agent    # Basic assistant
/dapr:agent durable my-agent      # Workflow-backed
/dapr:agent service my-agent      # REST API agent
/dapr:agent multi my-system       # Multi-agent

Related Resources

• Agent Templates: templates/agents/
• Pattern Templates: templates/agents/patterns/
• Tool Templates: templates/agents/tools/
• Integration Templates: templates/integrations/