| name | using-retrieval-augmented-generation-architect |
| description | Use the retrieval augmented generation architecture patterns neurosymbolic module: RAG library, agent brain, growable KB — proof-checked |
| version | 0.1.0 |
using-retrieval-augmented-generation-architect
This module is a CULTIVATED, PROOF-CHECKED knowledge organism about
retrieval augmented generation architecture patterns (207 concepts / 225 relations; grown by a
KB factory — every region admitted by a Prolog consistency gate, wrongness
tracked as open supersede-issues, never hidden).
The four ways to use it
- As RAG — the sibling
understand-{x} skills in this plugin,
coordinate-addressed (call number = home class : dependency facets — the
import web, literally); FTS5 index via skilltree.build_index over
${CLAUDE_PLUGIN_ROOT}/skills/using-retrieval-augmented-generation-architect/references/skilltree.json.
- As an agent (the runnable brain SHIPS here) — the neuromorphic brain
is bundled at
${CLAUDE_PLUGIN_ROOT}/skills/using-retrieval-augmented-generation-architect/data/brain/
(kuzu neurodb = the activation graph + tissue/ = the gyri). Point a
KbcBrain at it and brain_ask("your question"): the graph FIRES the
matching gyri numerically (spreading activation, weights decide — not an
agent choosing), each fired gyrus answers over its territory, the
synthesis is PROVEN one level up (the SES tower — the join is a theorem),
and the prover teaches the graph back (Hebbian). Needs kuzu +
brain-agent + ee_v2.kbc installed.
- As tools your agents hold —
ee_v2.kbc.heaven_tools.make_kbc_tools
over this module's data root
(${CLAUDE_PLUGIN_ROOT}/skills/using-retrieval-augmented-generation-architect/data/): 14 heaven tools
(kb_, kernel_, brain_*). Hand them to any heaven agent's tools=[...].
- As a factory — the kbworld round deepens this module on a schedule;
file a
kb-door issue to point it somewhere; file kb-supersede when
you catch it being wrong (it also catches itself — see the round reports).
The data resources (all inside THIS skill)
data/concepts.jsonl + data/relations.jsonl — the certified graphdata/hyperedges.jsonl — the certificate ledger (the automaton's KNOWN)data/skeletons.jsonl — certified argument DAGs (because/since/…)data/worklist.json — what the module knows it doesn't knowreferences/skilltree.json — the library index
Etiquette (the laws this module lives under)
- The prover admits; you never hand-edit certified state (file issues).
- Wrongness is fuel: a wrong-but-coherent region is a PENDING OBSERVATION —
say what you saw, the next round metabolizes it.
- The worklist is honest:
kb_work shows exactly what the module knows it
doesn't know.
The map
deep_retrieval_augmented_.memory_entryretrieval_augmented_generation_architecture_patt.retrieval_augmented_generation : deep_retrieval_augmented_(20)retrieval_augmented_generation_architecture_patt.dense_retrievalretrieval_augmented_generation_architecture_patt.vector_databaseretrieval_augmented_generation_architecture_patt.dense_passage_retrieverdeep_retrieval_augmented_.mmr001deep_retrieval_augmented_.mmr002deep_retrieval_augmented_.mmr003deep_retrieval_augmented_.mmr005deep_retrieval_augmented_.mmr009deep_retrieval_augmented_.mmr027deep_retrieval_augmented_.memory_store
