Ai dev
Use when building applications that call LLMs (Claude, OpenAI, Mistral, open-weights): prompt engineering, prompt caching, tool use / function calling, structured outputs, RAG, embeddings, vector stores, agentic workflows, evals, cost / latency optimization.From its SKILL.md
npx -y skills add PetrovC/ai-agent-kit --skill ai-devAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
1.2 KB, as published. Nobody here has run it
AI Development Skill
Goal
LLM-powered apps that are correct, cost-aware, observable, and testable. Treat the LLM like any other external dependency: typed inputs / outputs, retries, timeouts, evals, prompt versioning.
Quick reference
| Concept | Best practice |
|---|---|
| Caching | Order prompt stable -> volatile; cache tool definitions & system prompts |
| Outputs | Use tool calling / JSON mode with schemas for structured data |
| RAG | Chunk 500-1500 tokens, embed, store, retrieve top-k, and run reranking |
| Evals | Test with input -> expected behavior sets (50-500 cases), not snapshots |
| Latency | Cascade to smaller models (Haiku/3.5-mini), stream for chat UIs |
Full guidance
Extended how-to, patterns, anti-patterns, and checklists: SKILL.deep.md