using-llm-memory-and-context-engineering

name	using-llm-memory-and-context-engineering
description	Use the llm memory and context engineering neurosymbolic module: RAG library, agent brain, growable KB — proof-checked
version	0.1.0

using-llm-memory-and-context-engineering

This module is a CULTIVATED, PROOF-CHECKED knowledge organism about llm memory and context engineering (438 concepts / 213 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

1. 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-llm-memory-and-context-engineering/references/skilltree.json.
2. As an agent (the runnable brain SHIPS here) — the neuromorphic brain is bundled at ${CLAUDE_PLUGIN_ROOT}/skills/using-llm-memory-and-context-engineering/data/brain/ (kuzu neurodb = the activation graph + tissue/ = the gyri). Point a KbcBrain at it and brain_ask("your question"): the graph FIRES the matching gyri numerically (spreading activation, weights decide — not an agent choosing), each fired gyrus answers over its territory, the synthesis is PROVEN one level up (the SES tower — the join is a theorem), and the prover teaches the graph back (Hebbian). Needs kuzu + brain-agent + ee_v2.kbc installed.
3. As tools your agents hold — ee_v2.kbc.heaven_tools.make_kbc_tools over this module's data root (${CLAUDE_PLUGIN_ROOT}/skills/using-llm-memory-and-context-engineering/data/): 14 heaven tools (kb_, kernel_, brain_*). Hand them to any heaven agent's tools=[...].
4. 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 graph
• data/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 know
• references/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

• llm_memory_and_context_engineering.context_window : deep_context_window(13)
• llm_memory_and_context_engineering.context_length : deep_context_window(13)
• llm_memory_and_context_engineering.long_term_memory : deep_long_term_memory(14)
• llm_memory_and_context_engineering.working_memory : deep_context_window(13)
• llm_memory_and_context_engineering.attention_mechanism : deep_context_window(13)
• llm_memory_and_context_engineering.vector_database : deep_long_term_memory(2)
• llm_memory_and_context_engineering.conversation_history
• deep_long_term_memory.em_event
• llm_memory_and_context_engineering.retrieval_augmented_generation : deep_long_term_memory(14)
• llm_memory_and_context_engineering.sliding_window : deep_context_window(13)
• llm_memory_and_context_engineering.effective_context
• llm_memory_and_context_engineering.context_compression