agent-design

AI agent design and tool-use prompting patterns

⭐5stars🔀1forks📁View on GitHub🕐Updated Jan 11, 2026⬇️1downloads

When & Why to Use This Skill

This Claude skill provides a comprehensive framework for AI agent design, focusing on advanced tool-use prompting patterns and architectural blueprints. It enables developers to implement robust agentic workflows using industry-standard patterns like ReAct, Plan-Execute, and Reflexion, while offering structured templates for tool definitions and multi-tier memory management to enhance LLM reliability and task completion.

Use Cases

  • Implementing ReAct (Reasoning and Acting) patterns to create autonomous agents that can interact with external tools in real-time.
  • Designing complex, multi-step task solvers using Plan-Execute architectures to break down high-level objectives into manageable sub-tasks.
  • Configuring sophisticated memory systems (short-term, long-term, and episodic) to maintain context and improve agent performance over long conversations.
  • Standardizing tool definitions and error-handling schemas to prevent common agent failures like hallucinated parameters or infinite loops.
nameagent-design
descriptionAI agent design and tool-use prompting patterns
sasmp_version"1.3.0"
bonded_agent04-react-pattern-agent
bond_typePRIMARY_BOND

Agent Design Skill

Bonded to: react-pattern-agent


Quick Start

Skill("custom-plugin-prompt-engineering:agent-design")

Parameter Schema

parameters:
  agent_type:
    type: enum
    values: [react, plan_execute, reflexion, multi_agent]
    default: react

  memory_type:
    type: enum
    values: [none, short_term, long_term, episodic]
    default: short_term

  tool_count:
    type: integer
    range: [1, 20]
    default: 5

Agent Architectures

Architecture Strengths Use Case
ReAct Simple, effective General tasks
Plan-Execute Structured approach Complex multi-step
Reflexion Self-improvement Learning tasks
Multi-Agent Specialization Large systems

Core Patterns

ReAct Agent

## Agent Configuration
You are an AI assistant with access to tools.

## Available Tools
[Tool list with descriptions]

## Behavior
1. Think about what to do
2. Take an action using a tool
3. Observe the result
4. Repeat until task complete

Plan-Execute Agent

## Planning Phase
1. Analyze the task
2. Break into subtasks
3. Identify tools needed
4. Create execution plan

## Execution Phase
1. Execute each step
2. Validate results
3. Adjust if needed
4. Report completion

Tool Definition Template

tool:
  name: "tool_name"
  description: "When and why to use this tool"
  parameters:
    param1:
      type: string
      description: "What this parameter does"
      required: true
  returns: "Description of return value"
  errors:
    - "ERROR_TYPE: How to handle"

Memory Patterns

memory_types:
  working_memory:
    scope: current_conversation
    implementation: context_window

  long_term_memory:
    scope: persistent
    implementation: vector_store

  episodic_memory:
    scope: experience_based
    implementation: structured_logs

Troubleshooting

Issue Cause Solution
Wrong tool Vague descriptions Improve descriptions
Loops forever No exit condition Add max iterations
Forgets context Memory overflow Summarize periodically
Poor planning Complex task Add decomposition step

References

See agent frameworks: LangChain, AutoGen, CrewAI