rag-development
/rag-development
Zbiera wiedzę potrzebną do budowy, optymalizacji i audytu produkcyjnych systemów pytań i odpowiedzi RAG, w tym Graph RAG, CRAG, Self-RAG i Agentic RAG.
- Autor
- acaprino ↗
- Repozytorium
- alfio-claude-plugins ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 27.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 8
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 27.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Zbiera wiedzę potrzebną do budowy, optymalizacji i audytu produkcyjnych systemów pytań i odpowiedzi RAG, w tym Graph RAG, CRAG, Self-RAG i Agentic RAG.
Oryginalny opis (EN)
Knowledge base for production question-answering pipelines. TRIGGER WHEN: building, optimizing, or auditing RAG systems, including Graph RAG, CRAG, Self-RAG, and Agentic RAG.
Jak działa
- When to Use
- Quick Start Recommendation
- Reference Materials
- Pipeline Architecture
- Key Decision Points
Podgląd SKILL.md
--- name: rag-development description: > Knowledge base for production question-answering pipelines. TRIGGER WHEN: building, optimizing, or auditing RAG systems, including Graph RAG, CRAG, Self-RAG, and Agentic RAG. --- <daodan_host_binding host="opencode"> This binding governs execution of this package and its loaded references. It takes precedence over executable examples written for another host. Use only registered tools and their actual input schema; an example never creates a tool, argument, permission, model alias or nested worker capability. Preserve the generated worker allowlist, dependency closure, scope and delivery contracts. OpenCode V2 binding: repository reads use read/glob/grep, writes use write/edit/patch, shell execution uses shell, network tools are webfetch/websearch, and clarification uses the registered question tool. shell takes command, optional workdir, timeout in milliseconds and background. It does not provide interactive stdin. Persistent capture uses background: true and an explicit noninteractive stop signal. Load skills with skill using id: "<plugin>:<skill>". Dispatch registered roles with subagent using agent: "<owner>:<role>", description and prompt; optional sessionID continues that child, and background: true requests asynchronous completion. Use only roles allowed by the current harness and permissions. A child cannot dispatch nested workers when its permissions or the configured depth deny it. Workflow entries are /<plugin>:<workflow>, registered by the package loader. Inherit the session model; an override requires its authorized, actually available provider/model identity, not another host's short model alias. The loader includes required local providers and registers plugin-declared MCP servers. Qualified foreign providers need a compatible V2 plugin/skill provisioned through this host, never a Claude marketplace command. Check the active registry: registration metadata alone does not prove a connected server or usable tool. Inherit the active session's model and default variant. A source model alias such as sonnet is not a portable override; request a different model only through an explicit adapter mapping to a valid host identity and user authorization. The executing agent and its coordinator own continuation, recovery, verification and delivery. Skills supply knowledge and method instructions; they are not executors or a runtime persistence mechanism. Resolve each component by kind and exact installed identity. A role is dispatched as a worker, a skill loads instructions, and a workflow starts its declared method. Kernel owner-qualified skill references are logical provider identities: resolve them to the actual host registry name and verified installed provider path. Do not invent a registry string by adding an owner's prefix when the host does not use it. Do not load an agent identity as a skill or execute a workflow name as a shell command. Declared required foreign providers remain mandatory. If a required provider or capability (including required MCP access or isolation) is unavailable, stop its dependent phase and report the missing capability. Optional external providers follow the method's declared selection and absence behavior; their absence alone does not block the base workflow. Report any explicitly requested but unavailable provider contribution as unfulfilled and continue independent work. Do not install a different host's plugin or substitute unrelated skills to claim coverage. Shell execution alone does not establish interactive stdin. Before a capture or checkpoint, use an available session-control mechanism. When stdin is unavailable, keep user-driven capture in a supported background process with bounded output and an explicit stop signal controlled by the coordinator after the user's completion message. Close owned browser/process resources on completion or cancellation. Continue working within the current authorization while the requested result is achievable. After a tool error, inspect its cause, correct the relevant input or implementation, retry when justified and verify the outcome; never repeat a failing action blindly. Do not close the turn with a promise or a partial result when an authorized next step can still produce or verify the requested result. Stop only for a documented real impediment, a missing required capability, an unresolved human decision or the caller's stop request. At that boundary, report completed results, remaining work and the concrete prerequisite for continuation; continue any independent work that the blocker does not prevent. A tool's successful exit, a worker stop or an idle session is not the deliverable. Return final text naming observed findings, artifact paths, executed checks and blockers. An empty worker result is missing delivery: inspect or resume that worker when permitted, or record failure. Verify required artifacts and candidate gates before marking the requested outcome complete. This is an execution binding, not a rewrite of source examples. Preserve programming language fences, domain code and host APIs that are the subject of marketplace-ops or agent-sdk-builder. A host-as-subject API is not a tool provided to this session. Compiled binding and static checks do not prove an installed host executed a run. </daodan_host_binding> # RAG Development Comprehensive knowledge base for building production-grade Retrieval-Augmented Generation systems. ## When to Use - Building a new RAG pipeline from scratch - Choosing chunking strategy, embedding model, or vector database - Implementing hybrid search, re-ranking, or contextual retrieval - Evaluating RAG quality with RAGAS or DeepEval - Optimizing production RAG for cost, latency, or accuracy - Designing multi-tenant RAG with access control - Upgrading from naive RAG to advanced patterns ## Quick Start Recommendation For 80% of use cases, start with: 1. **Chunking**: Recursive character splitting at 512 tokens, 10-15% overlap 2. **Embedding**: OpenAI `text-embedding-3-small` (best value) or Cohere `embed-v4` (best accuracy) 3. **Vector DB**: Qdrant with scalar INT8 quantization 4. **Retrieval**: Hybrid search (dense + sparse + RRF) 5. **Evaluation**: RAGAS metrics from day one Then upgrade incrementally based on measured failures: - Keyword misses -> add sparse vectors (SPLADE/BM25) - Ambiguous chunks -> add contextual retrieval (Anthropic pattern) - Irrelevant results -> add cross-encoder re-ranking - Multi-hop failures -> upgrade to agentic RAG ## Reference Materials Detailed reference documents are in the `references/` directory: - `chunking-strategies.md` -- all chunking approaches with code, benchmarks, and selection guide - `embedding-models.md` -- model comparison, Matryoshka embeddings, fine-tuning, sparse/dense/multi-vector - `retrieval-patterns.md` -- hybrid search, HyDE, contextual retrieval, re-ranking, MMR - `advanced-rag-patterns.md` -- Graph RAG, RAPTOR, CRAG, Self-RAG, Agentic RAG, multi-modal RAG - `vector-databases.md` -- Qdrant deep dive, database comparison, scaling strategies - `production-guide.md` -- evaluation, observability, caching, security, cost optimization ## Pipeline Architecture ``` Document Ingestion: Raw Docs -> Preprocessing (Unstructured.io) -> Chunking -> Context Enrichment -> Embedding -> Vector DB Query Pipeline: User Query -> Query Transform -> Encode (Dense + Sparse) -> Hybrid Search -> Re-rank -> LLM Generation Evaluation Loop: Ground Truth + Predictions -> RAGAS/DeepEval -> Faithfulness, Relevancy, Precision, Recall ``` ## Key Decision Points | Decision | Default | Upgrade When | |----------|---------|-------------| | Chunking | Recursive 512 tok | Structured docs -> markdown-aware; cross-refs -> late chunking | | Embedding | text-embedding-3-small | Need accuracy -> embed-v4; self-hosted -> NV-Embed-v2 | | Vector DB | Qdrant + INT8 | Already on Postgres -> pgvector; need managed -> Pinecone | | Search | Dense only | Keyword misses -> add sparse hybrid; poor diversity -> add MMR | | Re-ranking | None | Top-k results contain irrelevant items -> add Cohere Rerank | | Caching | None | Production latency/cost concerns -> semantic cache | | Evaluation | Manual spot checks | Any production use -> RAGAS automated metrics |
Pliki w paczce (8)
- SKILL.md 8 KB
- SKILL.toml 1 KB
- references/advanced-rag-patterns.md 11 KB
- references/chunking-strategies.md 9 KB
- references/embedding-models.md 14 KB
- references/production-guide.md 11 KB
- references/retrieval-patterns.md 11 KB
- references/vector-databases.md 8 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · acaprino/alfio-claude-plugins | 100 | MIT |
| acaprino/alfio-claude-plugins | 100 | MIT |
| acaprino/alfio-claude-plugins | 100 | MIT |
| acaprino/alfio-claude-plugins | 100 | MIT |
| acaprino/alfio-claude-plugins | 100 | MIT |
| acaprino/alfio-claude-plugins | 100 | MIT |
Identyczna treść występuje jeszcze w 1 innych folderach na GitHubie. Tu pokazujemy jedną, najlepiej ocenioną kopię.
Migawka bazy z 13.11.2026 · wersja paczki 3d1d24a0e20a (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100agentic-rag
Opisuje wzorce RAG sterowane przez agenta: Self-RAG, Corrective RAG z fallbackiem na web, Adaptive RAG z klasyfikatorem routingu, ReAct z narzędziem retrieval, multi-hop i stany LangGraph. - 100corrective-rag
Wdraża Corrective RAG i Self-RAG, czyli niezawodne wyszukiwanie z samonaprawą, ocenia trafność dokumentów i dodaje samokorektę do potoków RAG. - 98langgraph-rag
Buduje maszyny stanów LangGraph dla RAG i przepływów agentowych: typowany stan, warunkowe krawędzie, checkpointing, przerwania human-in-the-loop, nadzorca wielu agentów oraz Self-RAG i CRAG. - 90agentic-rag
Pomaga budować samokorygujące się systemy wyszukiwania dla agentów AI: Corrective RAG, Self-RAG, dekompozycja zapytań, reranking, halucynacje i uziemienie. - 95production-rag-checklist
Wdraża systemy RAG na produkcję: architekturę, monitoring, skalowanie wyszukiwania i infrastrukturę operacyjną. - 100agentic-ai-dev
Dostarcza wzorce i szablony produkcyjnych agentów AI w Pythonie 3.14 z LangChain, LangGraph i FastAPI: agenci, systemy RAG, grafy przepływu, narzędzia, pamięć i testy.