메뉴
RAG pipeline production — ingestion, chunking chiến lược, embedding, hybrid search, rerank. Không index trùng, search dưới 200ms.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Expert in building 3D experiences for the web - Three.js, React Three Fiber, Spline, WebGL, and interactive 3D scenes. Covers product configurators, 3D portfolios, immersive websites, and bringing depth to web experiences. Use when: 3D website, three.js, WebGL, react three fiber, 3D experience.
Game art principles — art style, color theory, animation principles, asset pipeline
Game design principles — core loop, GDD, player psychology, difficulty balancing, progression
Orchestrator game — route to implementation skill (pxhopencode) + principle-based sub-skill (agent-skills-hub)
Mobile game principles — touch input, battery, thermal, app stores, monetization
Multiplayer game principles — networking architecture, synchronization, security, matchmaking
| name | ais-rag |
| description | RAG pipeline production — ingestion, chunking chiến lược, embedding, hybrid search, rerank. Không index trùng, search dưới 200ms. |
| File | Mục đích |
|---|---|
templates/chunking.py | chunk_document() — markdown/code/text splitters with overlap |
templates/embedding.py | EmbedCache (SQLite) + embed_batch() with OpenAI |
templates/schema.sql | pgvector DDL — documents table, IVFFlat + GIN indexes |
templates/hybrid-search.py | hybrid_search() — vector cosine + keyword ts_rank combined |
templates/rerank.py | LLM-based reranking with RERANK_PROMPT |
from templates.chunking import chunk_document
chunks = chunk_document(text, "markdown")
Performance targets: search < 200ms, rerank top-3 < 1s. Deduplication: hash content before insert to avoid duplicates.
| Excuse | Reality |
|---|---|
| "Hybrid search không cần, vector đủ" | Keyword miss → 0 results cho thuật ngữ chính xác |
| "Chunk overlap 0 là đủ" | Câu bị cắt giữa chừng → context incomplete |
| "pgvector index sau, data ít" | Full scan khi có 10k+ docs → search > 1s |