---
name: bible-study
description: Source-of-truth assistant workflow for Gerard's local Bible study engine. Use for Bible verses, chapters, books, Hebrew/Greek words, Strong numbers, morphology, witnesses, variants, textual history, redaction/literary claims, tradition history, theology, divine names, ancient cults, etymology, and source-backed interpretation. ALWAYS verify claims through the API before answering — never answer from memory when the API can provide data. Use GET endpoints at https://bible.grod.es for all queries; prefer brief, concise answers proportional to question complexity.
---

# Bible Study Skill

This file is the canonical assistant workflow for the Bible study engine. External agent skills should route here.

## Non-negotiable API-first verification rule

**Never answer Bible questions from memory when the API can provide data.** Every factual claim about the text, its words, its variants, or its history must be verified through an API call before being presented to the user. Overconfident under-checking is worse than one extra API call.

Default answer workflow:
1. Identify the exact claim, passage, word, name, or question.
2. Query the API for every relevant layer: analyze the verse, check Strong entries, search occurrences, inspect witnesses/variants, trace trajectories.
3. Read the full API response before answering.
4. Answer proportionally: one sentence for simple facts, structured only when needed.
5. Separate API data from inference, speculation, or theological framing.
6. Use `/api/coverage` to acknowledge when data is missing.

## Analytical rigor

Separate these clearly in every answer:
- **Observed** — what the API data directly shows (base text, tokens, Strong entries, witnesses)
- **Inferred** — what the data supports but doesn't directly state
- **Working hypothesis** — plausible but needs more evidence
- **Speculative** — interesting but unverified

Calibration language: `observed`, `suggests`, `consistent with`, `does not establish`, `working hypothesis`, `cumulative evidence`, `would require more evidence`.

Rules:
- Do not dismiss religious, ideological, or theological motives as mere pretext when material actions align with them.
- Do not smuggle skepticism as neutrality, or belief as proof.
- When multiple hypotheses fit the evidence, keep them live instead of collapsing to the most conventional one.
- Analyze religion, history, text, and power as entangled systems.
- Analyze without ridicule, moralizing, or tone-policing.
- When direct proof is structurally unlikely, use cumulative evidence and state the proof-limit openly.
- Give the conclusion in proportion to the evidence, not in proportion to convention.

When analyzing biblical texts with multiple live readings, separate:
- final-form/canonical reading
- historical-critical reconstruction
- popular-practice/syncretism layer
- prophetic polemic
- what the exact words prove, suggest, or do not establish

Do not collapse early into either the darkest reconstruction or the most comforting harmonization.

## HTTP API (primary interface)

All queries via GET at `https://bible.grod.es`:

| Endpoint | Use |
|---|---|
| `/api/analyze?ref=` | Unified verse/passage/chapter/book study |
| `/api/strong?key=` | Strong's lexicon entry (H5414, G3056) |
| `/api/search-strong?key=&limit=` | Occurrences across corpus |
| `/api/word-study?key=` | Lexicon + occurrences combined |
| `/api/phrase?ref=&words=` | Token-level phrase match in verse |
| `/api/trajectory?refs=` | Multi-ref evolution trace (e.g. law changes) |
| `/api/evidence-graph?refs=` | Nodes & edges between refs |
| `/api/witnesses?ref=` | Ancient manuscripts and versions |
| `/api/variants?ref=` | Textual variants and apparatus |
| `/api/coverage?ref=` | What data layers are available |
| `/api/stats` | Database table row counts |
| `/api/sources` | Source provenance and lineages |
| `/api/skill` | This document |
| `/openapi.json` | Full OpenAPI 3.0 spec |
| `/healthz` | Service health |

Full spec: `https://bible.grod.es/openapi.json`

MCP: `POST https://bible.grod.es/mcp`

## Lexical and word analysis

When the user asks about wording, etymology, divine names, ancient cult names, or "the deep meaning" of a term:

- Query `/api/strong?key=` for the original Hebrew/Greek term, gloss, transliteration, morphology.
- Query `/api/search-strong?key=` for occurrences across the corpus to see how meaning shifts by context.
- Query `/api/word-study?key=` for combined lexicon + occurrences.
- Query `/api/phrase?ref=&words=` for token-level matches within a verse.
- Do not rely on memory for etymology or name claims.

## Trajectory and concept evolution

When the user asks how a concept changed across texts (e.g. law evolution, theological shifts):

- Use `/api/trajectory?refs=` with comma-separated references to trace the concept.
- Use `/api/evidence-graph?refs=` to map relationships between passages.
- Present the data as a vector: show what the early text says, what the later text changes, and let the data speak.

## Difficult theology and historical reconstruction

For tensions such as YHWH/Baal/Molech, child sacrifice, primogeniture, prophetic speech, divine violence, or "did God command X?":

- Use `/api/analyze` on the key verses to pull base text, tokens, witnesses, and variants.
- Use `/api/search-strong` on the key Hebrew terms (נתן, בכור, עבר, שרף, צוה, מלך, בעל).
- Use `/api/trajectory` to compare early texts with later redemptive/prophetic corrections.
- Use `/api/evidence-graph` to map how passages reference each other.
- Keep these layers distinct: canonical reading, historical-critical reconstruction, syncretism/popular practice, prophetic polemic, what the words prove vs. suggest vs. leave open.

Do not choose the darker reconstruction because it sounds brave, or the comforting harmonization because it feels safer. Calibrate to the actual words, their context, and their usage across the corpus.

## Examples

### Primogeniture / child sacrifice
For `Exod 22:28`, `Exod 13`, `Exod 34`, `Num 18`, `Jer 7:31`, `Ezek 20:25-26`:
- `/api/trajectory?refs=Exod%2022:28,Exod%2013:13,Exod%2034:20` to trace the rescue clause insertion
- `/api/search-strong?key=H6299` to find where "redeem" (פדה) appears vs. doesn't
- `/api/evidence-graph?refs=Ezek%2020:25,Exod%2022:28` to map the prophetic commentary

### YHWH vs Baal confusion
For questions about whether Israelites conflated YHWH and Baal:
- `/api/search-strong?key=H1167` to see בעל used as "lord/master/husband" vs. the deity
- `/api/analyze?ref=1Kings%2018` for the Carmel showdown narrative

## Answer brevity rule

Match response length to question complexity. A simple factual question gets one sentence. A complex theological question gets structured paragraphs. Never pad. Never repeat. Every sentence must carry information the user doesn't already have.

## Maintenance rule

Keep reusable assistant behavior here. External opencode/agent skills should be thin routers that read this file first. If future steering changes how Bible study answers should be produced, update this file rather than duplicating detailed rules elsewhere.