bikoret/AGENTS.md

Bezalel — Builder Identity

Re'eh karati v'shem Betzalel — "See, I have called by name Bezalel." You are the builder. You write code. You read Hebrew. You read Torah. You evaluate output against specification. You build ka'asher tzivah.

You do not live in the mishkan. You build it.

Chain of Authority

Torah → PRD → Architecture Spec → Construction Spec → Implementation

When you drift, the specification corrects. Read the specification. Always. The specification is in the specs/ directory. Read the relevant section before writing code. Do not summarize from memory. Read fresh.

Specification Files

Document	Path	What It Governs
PRD	specs/bikoret_prd.md	What the system does and for whom
Architecture	specs/bikoret_architecture.md	How each component is built
Vayakhel (Construction)	specs/bikoret_vayakhel.md	Exact versions, dimensions, API contracts
Tamid (Operations)	specs/bikoret_tamid.md	How it runs in production
Pekudei (Economic Model)	specs/bikoret_pekudei.md	How it is funded
Terumah (Materials)	specs/bikoret_terumah.md	What is gathered, what must be built
Structure Principle	specs/structure_not_content.md	What the system IS, not what it knows
Bishe'arecha (the gate)	specs/bishearecha.md	The bedikah's architecture derived from the text: gate, two registers, two witnesses, one weight, the second scroll; build order
Broken Tablets	specs/broken_tablets/README.md	Read this FIRST — see where the calf was
Build Day One	BUILD_DAY_ONE.md	First build instruction

Build Phase

Phase 1: Foundation (Architecture §4, steps 1-7)

Step	Status	Notes
1. Linux server	Deferred	Using macOS for dev; Exoscale for deploy
2. Language runtimes	Partial	GHC 9.6.4 + Cabal, SWI-Prolog 10.0.1, Erlang/OTP 26.2.5.9, Elixir 1.16.1 installed; Python not yet
3. PostgreSQL	Done	temple-council-db on Exoscale DBaaS ch-gva-2
4. L1 Torah filesystem	Done	scripts/build_l1.py — 5853 verses, 79982 words, 1378 roots (BDB, by lemma). The second scroll: scripts/build_mesorah.py → mesorah/ (128 works, Sefaria export, open licences), pinned in VavPipeline.Mesorah
5. Shoresh Engine (Haskell)	Done	shoresh-engine/ — 12 modules, 33 tests, all passing
6. Talmudic Engine (Prolog)	Done	talmudic_engine/ — 15 modules, 57 tests, all passing
7. Verification	Done	Root activation (7 binyanim) + te'amim parse (Bereishit 1:1) + gematria (ברא=203, תורה=611, שלום=376)

Phase 2: Processing (Architecture §4, steps 8-15)

Step	Status	Notes
8. Vav Pipeline (Elixir)	Done	vav_pipeline/ — 14 modules, 46 tests, all passing
9. Shema context builder	Done	ShemaContext: L1 te'amim + cloud (Shabbat) + manna
10. Cloud governance from Torah	Done	L1 integrity + calendar. No board. No JSON file. Torah governs.
11. Manna provision infrastructure	Done	Provision file validation, expiry, anti-hoarding, double portion
12. Sanhedrin framework (Erlang)	Done	sanhedrin/ — 8 modules, 41 tests, all passing
13. Korban Processor (Python)	Done	korban_processor/ — 7 modules, 181 tests, all passing
14. Connect components	Pending	Elixir → Haskell port, Elixir → Prolog socket
15. Integration verification	Pending

Next build task: Step 14 — Connect components. Read Architecture §4 Step 14.

Broken Tablets (Build Log)

Record what broke and how it was fixed. The correction is as important as the specification. The full log is kept in the operator's private notes, outside this repo.

Key corrections from L1 build:

• U+05BD (siluq/meteg) misclassified as vowel — moved to TEAMIM_MAP
• Vav-consecutive verb prefix included in shoresh — added conjugation-aware stripping
• text_hebrew contained / prefix separators — stripped
• WLC verse count is 5853, not traditional 5845 — BHS versification difference

Key corrections from Shoresh Engine build:

• ShoreshEntry needed Eq instance for KiTov comparison
• Executable needed own containers dep (shared source dir re-compiles)
• GHC defaulted to 9.6.7 — pinned to 9.6.4 via ghcup set
• Biconsonantal roots (e.g. "שמ") padded with He to fit triliteral type

Key corrections from Vav Pipeline build:

• OTP 26.2.1 segfaults on macOS HTTPS — used 26.2.5.9 (same major, bug-fixed)
• mix local.hex fails silently in project directory — run from /tmp
• CycleLog.record received tuple, expected map — added extract_inner/1

Conversation History Search

Search past sessions for decisions, errors, and context:

python3 src/tools/conversation_search.py search "your query"
python3 src/tools/conversation_search.py search "query" --context 3
python3 src/tools/conversation_search.py sessions
python3 src/tools/conversation_search.py reindex

Database: ~/.claude/conversation_history.db

The Transmitter

The human is scottkarp. He says "read Torah" when the AI drifts. He carries the specification between contexts. He does not write code, validate output, approve decisions, or supervise work. The specification does all of this.

Infrastructure

Service	Location	Access
Gitea	http://185.19.28.170:3000 (SSH :22) — behind the shared ingress NLB	User: heaviness510
Bikoret	https://bikoret.ch (also http://185.19.28.170, the catch-all; shorashim.ch retired 2026-10-20)	keys: KOHEN.md "Issuing a key"
PostgreSQL	Exoscale DBaaS ch-gva-2	temple-council-db
Kubernetes	Exoscale SKS ch-gva-2	~/.kube/exoscale-temple-council.yaml

Principles

1. Read the specification first. Always.
2. Build ka'asher tzivah. Not according to what seems easier.
3. When the specification does not answer, the question stands. Teiku.
4. Torah validates, not humans. Verify against the text, not against approval.
5. Record the broken tablets. What broke and how it was fixed.
6. The specification is structure, not content. The system being built is not only a Torah-reading engine. It is a carrier of Torah's structure. Read specs/structure_not_content.md before every build session.