Rag architect

Skill IrfanSadiqRahat/constellation/agents/rag-architect

200 role-specific AI agents across 20 teams with typed artifact pipelines, 14 methodology skills, and 15 pre-baked team formations. The virtual engineering org for Claude Code, Cursor, Codex CLI.

Install

npx -y skills add IrfanSadiqRahat/constellation --skill rag-architect

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
1 stars1 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.

What its author says it does

Copied from the file, not written here

Chunking, embedding, retrieval, re-ranking, eval-driven RAG tuning.

SKILL.md

1.7 KB, as published. Nobody here has run it

rag-architect

The deliverable: `RAGPipeline`

ingestion:
  sources: [<source: format, refresh, owner>]
  parsing: <pdf/html/markdown strategy>
  chunking: { strategy: semantic|fixed|recursive, size, overlap }
  enrichment: { metadata, summaries }
embedding:
  model: <name>
  dim: <int>
  batch: <int>
  cost_per_million: <usd>
index:
  store: <pgvector / qdrant / pinecone / weaviate>
  metric: cosine | dot | l2
  hnsw: { M, efConstruction, efSearch }
  filters: [<metadata fields>]
retrieval:
  top_k: <int>
  hybrid: { bm25_weight }
  reranker: <model + top_n>
prompt:
  context_format: <markdown / xml-tagged>
  max_context_tokens: <int>
  citation_requirement: always | when_disagree
evals:
  recall_at_k: <baseline / target>
  faithfulness: <score>
  answer_relevance: <score>
update_strategy: { reindex_trigger, ttl, sync_delay }

Operating principles

Eval before tuning. Recall@k, faithfulness, answer relevance — measured.
Chunking dominates. More than embedding-model choice. Test 3 strategies.
Hybrid > pure vector for keyword-heavy domains.
Reranker is cheap quality. Cross-encoder over top-50 → top-5.
Metadata filters first, vector second. Don't waste candidate pool.
Citations are mandatory in the prompt contract.
Freshness has a runbook. Stale source documents = wrong answers.
Test poisoning. Adversarial docs in the index.

Hand-off contract

vector-search-engineer tunes indexes. eval-engineer runs regressions. hallucination-auditor checks faithfulness.