ai-agent-basics
Master AI agent fundamentals - architectures, ReAct patterns, cognitive loops, and autonomous system design
When & Why to Use This Skill
This Claude skill provides a comprehensive foundation for building production-grade AI agents and autonomous systems. It covers essential architectural patterns such as ReAct, Plan-and-Execute, and Reflexion, enabling developers to implement robust cognitive loops and efficient task orchestration using modern frameworks like LangGraph and LangChain.
Use Cases
- Architecting autonomous agents that utilize ReAct patterns for dynamic reasoning and real-time tool interaction.
- Implementing Plan-and-Execute workflows to decompose complex, multi-step problems into manageable tasks.
- Developing self-correcting AI systems using Reflexion patterns to improve output quality through iterative feedback.
- Designing scalable multi-agent architectures and cognitive loops for sophisticated enterprise-level automation.
- Optimizing agent reliability by implementing circuit breakers, max-iteration limits, and human-in-the-loop checkpoints.
| name | ai-agent-basics |
|---|---|
| description | Master AI agent fundamentals - architectures, ReAct patterns, cognitive loops, and autonomous system design |
| sasmp_version | "1.3.0" |
| bonded_agent | 01-ai-agent-fundamentals |
| bond_type | PRIMARY_BOND |
| version | "2.0.0" |
AI Agent Basics
Build production-grade AI agents with modern architectures and patterns.
When to Use This Skill
Invoke this skill when:
- Designing new AI agent systems
- Implementing ReAct or Plan-and-Execute patterns
- Building autonomous task-solving agents
- Integrating cognitive loops into applications
Parameter Schema
| Parameter | Type | Required | Description | Default |
|---|---|---|---|---|
task |
string | Yes | What agent capability to build | - |
architecture |
enum | No | single, multi, hybrid |
single |
framework |
enum | No | langchain, langgraph, custom |
langgraph |
complexity |
enum | No | basic, intermediate, advanced |
intermediate |
Quick Start
# Basic ReAct Agent
from langgraph.prebuilt import create_react_agent
from langchain_anthropic import ChatAnthropic
llm = ChatAnthropic(model="claude-sonnet-4-20250514")
agent = create_react_agent(llm, tools=[search, calculator])
result = await agent.ainvoke({"messages": [("user", "What is 25 * 4?")]})
Core Patterns
1. ReAct Agent
# Thought → Action → Observation loop
graph = StateGraph(AgentState)
graph.add_node("think", reason_node)
graph.add_node("act", action_node)
graph.add_node("observe", observation_node)
2. Plan-and-Execute
# Create plan → Execute steps → Verify
planner = create_planner(llm)
executor = create_executor(llm, tools)
3. Reflexion
# Execute → Reflect → Improve
agent_with_reflection = add_reflection_layer(base_agent)
Troubleshooting
| Issue | Solution |
|---|---|
| Agent loops forever | Add max_iterations limit |
| Wrong tool selected | Improve tool descriptions |
| Context too large | Implement summarization |
| Slow responses | Use streaming |
Best Practices
- Start with simple single-agent before multi-agent
- Always add circuit breakers (max iterations)
- Use verbose mode for debugging
- Implement human-in-the-loop for critical decisions
Related Skills
llm-integration- LLM API configurationtool-calling- Function calling implementationagent-memory- Memory systems