Guide creation of Claude Code agents with architecture and prompt engineering.
npx skills add https://github.com/agdev/claude-code --skill agent-creator-agdev
Please help me install this Agent Skill. Skill: agent-creator Source: https://github.com/agdev/claude-code/tree/main/skills/agent-creator Command: npx skills add https://github.com/agdev/claude-code --skill agent-creator-agdev
Designing effective AI agents (subagents) that perform complex tasks autonomously requires specialized knowledge in architecture, prompt engineering, and tool selection. This Skill provides the blueprint to create agents that work, ensuring they are focused, efficient, and secure.
code-reviewer agent that automatically analyzes pull requests for security, performance, and maintainability issues, providing structured, actionable feedback directly within your workflow.Create a new Claude Code agent named documentation-writer that specializes in generating API documentation from source code.
High-intent search queries and answers about installing and using this skill.
Create an AI agent by defining its single-purpose focus, crafting a clear system prompt with concrete examples, selecting appropriate tools and Claude model (Haiku, Sonnet, or Opus), and structuring configuration files following agent architecture best practices for autonomous task execution.
Effective agent prompts use strong emphasis markers, concrete examples, and clear behavioral instructions. Structure prompts to specify the agent's role, constraints, tool usage patterns, and expected output format to ensure focused, reliable autonomous performance.
Choose based on task complexity and cost trade-offs: Haiku for simple, high-volume tasks; Sonnet for balanced performance and cost; Opus for complex reasoning. Match model capability to your agent's autonomy requirements and security constraints.
Single-purpose agents are more focused, efficient, and secure. They reduce decision overhead, minimize tool misuse risk, simplify testing and debugging, and perform their designated task more reliably than generalist agents handling multiple workflows.
Yes, subagents can be composed into larger agent systems. Design each subagent with single-responsibility focus, validate tool access constraints between agents, and document standardized handoff patterns to maintain security and predictability.
bsamieeCreate Claude Code agent files with YAML frontmatter and tool permissions.
bmsull560Evaluate agent performance and extraction accuracy in LangGraph pipelines.
vecearReduce LLM costs and latency via caching, model selection, batching, and prompt optimization.
handsomelong922Automate building, deployment, and management of autonomous AI agents.