AgentSkillsCN

dive-coder-rag-enterprise

企业级 RAG v2（离线优先）：增量索引、BM25 + 混合钩子、GraphRAG/RAPTOR/CRAG 适配器、评估与 EvidencePack。

SKILL.md
--- frontmatter
name: dive-coder-rag-enterprise
description: "Enterprise RAG v2 (offline-first): incremental indexing, BM25 + hybrid hooks, GraphRAG/RAPTOR/CRAG adapters, eval + EvidencePack."
metadata: '{"version":"14.4.0","project":"Dive Coder","tags":["rag","bm25","hybrid","graphrag","raptor","crag","eval","evidencepack"],"offline_first":true}'

What this skill does

  • •RAG v2 core: incremental ingest + hash cache, BM25 lexical retrieval, heuristic multi-query + step-back, overlap rerank, max context cap
  • •Advanced adapters (skeletons): GraphRAG / RAPTOR / CRAG pipelines (offline-first defaults)
  • •Governance outputs: report JSON, claims ledger (E2), EvidencePack (E3 bundle)

CLI usage (repo root)

bash
export PYTHONPATH="$PWD/.shared/vibe-coder-v13:${PYTHONPATH:-}"

# Build / update KB (skips unchanged docs)
python3 .shared/vibe-coder-v13/vibe.py rag ingest \
  --spec .vibe/inputs/v13/rag_spec.yml \
  --kb .vibe/kb

# Query
python3 .shared/vibe-coder-v13/vibe.py rag query \
  --kb .vibe/kb \
  --q "your question" \
  --max-context-chars 24000

# Eval (report + claims + evidencepack)
python3 .shared/vibe-coder-v13/vibe.py rag eval \
  --kb .vibe/kb \
  --eval .vibe/inputs/v13/rag_eval.yml \
  --out .vibe/reports/rag-eval.json \
  --evidencepack-out .vibe/evidencepacks

Where to extend

  • •Dense embeddings + hybrid fusion: .shared/vibe-coder-v13/rag/adapters/embedding/*
  • •Cross-encoder / LLM rerank: .shared/vibe-coder-v13/rag/adapters/rerank/*
  • •GraphRAG/RAPTOR/CRAG: .shared/vibe-coder-v13/rag/advanced/*