LLM 與 Agent

Claude Code 스킬 생성 및 관리. 스킬 구조, 트리거, 훅, skill-rules.json 작업을 다룹니다.

Decompose PRDs and Tech Specs into parallel-executable tasks with contracts, prompts, and dependency graphs. Use when breaking down a PRD for multi-agent execution.

Breaks down requirements into iterations and tasks

외부 AI CLI(codex, qwen, copilot, rovo-dev, aider)와 Claude의 협업 루프. 도구와 역할을 선택하여 계획-구현-리뷰 사이클을 실행. 기본값은 codex CLI.

Run automatically after completing any code changes. Validates TypeScript build, auto-fixes prettier and linting, then manually fixes any remaining issues.

Handles camera integration, photo capture, sticker creation, and storage workflows in Expo React Native apps. Use when implementing camera features, photo editing (sticker creation), file system operations, and integration with MST.

Log migration details to MIGRATION_LOG.md only. Use after completing a file migration to record details, issues, and solutions. Does NOT modify CLAUDE.md or TASK_BOARD.md - those are managed by other skills. Focused solely on maintaining detailed migration history.

Systematic PRD-driven software development workflow for Claude Code. Use when the user wants to (1) Import/parse feature bundles into draft PRDs with dependency analysis, (2) Create comprehensive Product Requirements Documents from feature descriptions, (2b) Extend existing PRDs with new requirements (version increment), (3) Break down PRDs into structured task lists with dependencies and test plans, (4) Systematically execute task lists with quality gates and verification protocols, (5) Audit test coverage and correctness against specifications, (6) Generate comprehensive PRD status reports showing completion states and progress tracking, or build complete applications from requirements through to tested implementation following enterprise-grade practices.

A comprehensive, long-horizon skill for designing, coding, debugging, and scaling AI-integrated systems — including Python backends, React frontends, RAG pipelines, vector databases, and tool orchestration. Ensures production-grade quality and consistency across every code and architecture decision.

Manages ticket CRUD operations for the project management system. Handles creating, reading, updating, listing, and searching tickets stored in .claude/data/tickets/.

Production-grade agentic system development with LlamaIndex in Python. Covers semantic ingestion (SemanticSplitterNodeParser, CodeSplitter, IngestionPipeline), retrieval strategies (BM25Retriever, hybrid search, alpha weighting), PropertyGraphIndex with graph stores (Neo4j), context RAG (RouterQueryEngine, SubQuestionQueryEngine, LLMRerank), agentic orchestration (ReAct, Workflows, FunctionTool), and observability (Arize Phoenix). Use when asked to "build a LlamaIndex agent", "set up semantic chunking", "index source code", "implement hybrid search", "create a knowledge graph with LlamaIndex", "implement query routing", "debug RAG pipeline", "add Phoenix observability", or "create an event-driven workflow". Triggers on "PropertyGraphIndex", "SemanticSplitterNodeParser", "CodeSplitter", "BM25Retriever", "hybrid search", "ReAct agent", "Workflow pattern", "LLMRerank", "Text-to-Cypher".

Use this skill when the user asks to create a new skill, custom command, or wants to know how skills work in Claude Code.

Publishing assistant for Thomas's Astro-based GitHub Pages site (readme-pages). Creates publish-ready Markdown files for notes, recipes, and pages. Use when asked to write blog posts, articles, notes, recipes, or content for the readme-pages site at thomasrohde.github.io/readme-pages. Also use for content requests mentioning "publish", "blog", or "my site".

Generate professional prompts for Nano Banana Pro image generation. Use when the user wants to create images, thumbnails, infographics, edit photos, generate character-consistent content, create storyboards, or any visual asset using Nano Banana Pro. Transforms vague requests into detailed, effective prompts following Nano Banana Pro's "thinking model" best practices.

Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.

OpenAI official SDK usage (Python, Node.js). Use when: writing code that calls OpenAI API,implementing chat/embeddings/images/audio features, handling streaming responses,async patterns, error handling with SDK. For raw HTTP/REST calls, see `openai-api` skill.

Master QMIX, MADDPG, CTDE - multi-agent learning with coordination and credit assignment

Use when multiple independent tasks can run simultaneously. Enables efficient parallel work execution with specialized agents.

LLM integration patterns for Claude, GPT, Gemini, and Ollama. Activate for AI API integration, prompt engineering, token management, extended thinking, and multi-model orchestration.

Programmatic verification of KB state: tickets, phases, tests, integrity.Provides scripts for automated checking at each workflow stage.SCRIPTS:- check_ticket.py: Ticket docs, tests, final status → next step- check_phase.py: Phase progress, all tickets complete?- check_tests.py: Trajectory, RED verification, coverage vs spec- check_integrity.py: Index.md matches dirs, roadmap has all active itemsUse when:- Checking ticket/phase completion- Validating before commit/archive- CI/CD automation- User says "status", "check", "verify"