AI Agent Helper

**Quick judgment** AI Agent Helper is a focused, prompt-engineering–oriented skill for people who are building or tuning AI agents. It appears most useful for users who already have a basic understanding of agents and want help with system prompt design, task decomposition, ReAct/CoT-style loops, tool selection, and structured outputs. **Strengths** - Good conceptual coverage of key agent-design pieces: system prompts, task decomposition, loop design, tool usage, and token optimization. - Provides concrete structural patterns (role/goal/constraints/output format, JSON/structured outputs, few-shot examples), which are directly actionable for production-style agents. - Especially valuable for Cantonese/Chinese-speaking builders, as the descriptions and examples are tailored to that audience. **Risks / Limitations** - No explicit mention of safety-by-design patterns (e.g., guardrails, abuse-prevention prompts, alignment constraints), so the skill may help optimize agents without adequately addressing misuse or harmful outputs. - Quality and consistency depend heavily on how well the user frames their goals; beginners may still struggle to specify requirements clearly. - The description doesn’t indicate deep integration with specific frameworks (LangChain, OpenAI Assistants, etc.), so guidance may remain somewhat generic. **Recommended scenarios** - Builders who want to quickly iterate on prompt structures, agent goals, and constraints for task-focused agents. - Developers integrating LLM agents into apps and needing better JSON/structured output design and error-handling patterns. - Intermediate users who already know what an "agent" is but need help making it more reliable, cheaper (token-wise), and better aligned with the desired task behavior.