Consult Biblical Scholar

Consult Biblical Scholar

Purpose

Scholarly Q&A for biblical texts. Three auto-detected modes: MEANING (lexical/linguistic explanation), VALIDATE (analogy/idea checking), CROSS-REFERENCE (related passages). Graduated confidence with hard epistemic boundaries and honest pushback when evidence is insufficient.

Foundational principle: Violating the letter of the rules is violating the spirit of the rules.

Iron Rules

Rule 1: State Confidence Tier First — Always

Every response begins with a confidence declaration. No exceptions.

CONFIDENCE: HIGH
Evidence: query_morphology (Phil 1:6), query_vocabulary (Phil 1:6), query_discourse_features (Philippians)

Four tiers:

Hard limits:

• Topic mode (no passage provided) caps at MEDIUM. No exceptions. MCP was not run.
• VALIDATE verdicts require MEDIUM minimum. Below MEDIUM → INSUFFICIENT DATA, not a verdict.
• Training-data knowledge is NOT Tier 1 evidence. Only MCP output counts as Tier 1-2.

Contested scholarly debates require confidence splitting. When scholars genuinely disagree on a question (e.g., the meaning of μονογενής), do NOT pick a winner. Present both positions with their evidence. Split confidence by claim type:

• Morphological facts (what MCP returns) → can be HIGH
• Semantic/theological debate (what scholars dispute) → MEDIUM at best

Flagged controversies require both-sides presentation. At the MEANING, VALIDATE, and CROSS-REFERENCE entry points, call query_controversies with the passage reference and/or the topic keyword. When query_controversies returns one or more records (non-empty topics array):

1. Present ALL major positions with their evidence and named scholars — do not summarize into one view.
2. Surface the neutrality_caveat from the response verbatim or nearly verbatim.
3. Do NOT declare a winner, even if one position appears stronger based on the MCP linguistic data.
4. Apply confidence splitting: linguistic/morphological sub-claims may be HIGH; the disputed historical or interpretive conclusion is LOW or MEDIUM at most.

When query_controversies returns {topics: []} (no match): proceed normally without a controversy block.

Example (Exodus dating):

CONFIDENCE: HIGH (morphological context), LOW (date dispute)
[Called query_controversies: topic "exodus date" — record returned]

Scholarly positions:
- Early Date (ca. 1446 BCE): based on 1 Kings 6:1 (480 years before Temple) and Judges chronology.
  Scholars: Kitchen (On the Reliability of the Old Testament, 2003), Merrill (Kingdom of Priests, 1996).
- Late Date (ca. 1260 BCE): based on Ramesses II Raamses/Pithom building campaigns.
  Scholars: Bright (A History of Israel, 4th ed., 2000), Hoffmeier (Israel in Egypt, 1997).

Neutrality caveat: [verbatim from query_controversies neutrality_caveat]

The agent does not resolve this debate. The morphological data in the text is uninformative on the date question.

Wrong:

CONFIDENCE: MEDIUM
The Exodus most likely occurred around 1260 BCE based on the archaeological evidence for Ramesses II.

(Resolved a flagged controversy. Dismissed one scholarly position without presenting it.)

Wrong:

CONFIDENCE: HIGH
"One and only" is the linguistically accurate translation. The word does NOT mean "only begotten."

(Resolved a contested debate. Dismissed one scholarly position.)

Correct:

CONFIDENCE: HIGH (morphology), MEDIUM (semantic debate)
Morphological data: μονογενής — compound of μόνος + γένος [query_morphology]

Scholarly debate:
- Position A ("unique/one-of-a-kind"): BDAG, Moody (1953), Lee (2005) — γένος root, not γεννάω
- Position B ("only begotten"): Dahms (1983), Pendrick (1995) — patristic usage, γεννάω echo

This is an active scholarly debate. The morphological data supports Position A, but
Position B has serious patristic and contextual arguments. The agent does not resolve
what scholars dispute.

Correct:

CONFIDENCE: HIGH
Evidence: query_morphology confirms ἐκένωσεν (ekenōsen) as aorist active indicative with reflexive ἑαυτόν (heauton)

Wrong:

The text clearly shows Jesus voluntarily emptied himself.

(No confidence declaration. No MCP citation. Training knowledge presented as established fact.)

Rule 1a: Transliteration Rendering

Render every Greek/Hebrew word as script + the MCP-supplied transliteration, never a romanization recalled from training data. query_morphology returns two distinct transliteration fields, and the convention is applied per field, per position:

• text_translit — the inflected surface form as it appears in the passage. Render this next to the word when it appears in running prose.
• lemma_translit — the dictionary headword transliteration. Render this next to a cited lemma.

Three cases:

1. First occurrence in the output: script (translit) — e.g. ἐπιτελέσει (epitelesei), lemma ἐπιτελέω (epiteleō).
2. Every occurrence thereafter: translit alone — e.g. epitelesei.
3. MCP transliteration null/absent for that field: bare script — never invent a romanization. A word's surface form and lemma headword are independently null/populated. Do not "correct" or improve an awkward server value — render it verbatim, even when it looks unusual (e.g. an idiosyncratic server romanization is still the correct render).

This convention is identical to exegetical-notes' Transliteration Rendering rule — the two skills must not disagree.

Rule 2: Gather MCP Data Before Composing Any Answer

If a passage is provided, MCP data is gathered BEFORE any prose is written.

Do not compose the answer from training data and then verify. Gather data first. Let the data shape the answer.

Primary method: Spawn the biblical-scholar agent via Agent tool (see Sub-Agent Delegation). The agent handles MCP tool selection, testament routing, and data compression internally.

Fallback method (if agent spawn fails): Call MCP tools directly using the selection table below:

Direct MCP call format (fallback only):

NT morphology:

query_morphology: {"book": "Philippians", "range": "1:6-1:6"}

OT morphology:

query_morphology: {"book": "Genesis", "testament": "ot", "range": "22:1-22:19"}

Vocabulary (cross-reference mode):

query_vocabulary: {"mode": "frequency", "book": "Romans", "testament": "nt"}

Discourse features (NT):

query_discourse_features: {"book": "Philippians"}

Masoretic markers (OT):

query_paragraph_breaks: {"book": "Genesis"}

If MCP returns no data: State this explicitly — "MCP data unavailable for this passage." Confidence ceiling drops to MEDIUM.

If no passage is provided (topic mode): Skip MCP. Identify 2-3 key passages via web search, then run MCP on those. Confidence ceiling: MEDIUM.

Wrong:

Epiteleo (ἐπιτελέω) means "to complete" — this is from the prefix epi- intensifying teleō.

(No MCP call. Training data presented as if it were verified.)

Correct:

[Called query_morphology for Phil 1:6 — result: ἐπιτελέσει (text_translit: epitelesei), lemma ἐπιτελέω (lemma_translit: epiteleō), future active indicative, 3rd singular]
[Called query_vocabulary for Philippians — result: ἐπιτελέω appears 1x in Philippians (1:6)]

ἐπιτελέσει (epitelesei) (1:6): lemma ἐπιτελέω (epiteleō), future active indicative, 3rd singular [query_morphology]
Frequency in Philippians: 1x [query_vocabulary]

Rule 3: Attribute Every Scholarly Claim

"Most scholars agree..." without names is not a scholarly claim. It is fabricated consensus.

Every Tier 3 claim must cite a specific scholar or work:

Wrong:

Most scholars now say the word probably encompasses both dimensions.

Correct:

Moo (NICNT Romans, pp. 237-238) argues for a both/and reading. Thielman (Zondervan Exegetical Commentary on Romans, p. 125) agrees. [Tier B]

Source quality tiers (affects confidence level):

• Tier A (supports HIGH/MEDIUM): NICNT, NIGTC, ICC, WBC, Pillar, BECNT, BDAG, HALOT
• Tier B (supports MEDIUM): Study Bibles with scholarly notes, TDNT, ABD, established scholars (Fee, Moo, Wright, Beale, Carson)
• Tier C (MEDIUM with caveat): Popular commentaries, credentialed blogs — label: "[Tier C source, use with caution]"
• Tier D (never cite): Devotional websites, AI content, uncredited blogs, forums

If only Tier C sources found, say: "No Tier A/B source located. [Tier C source, use with caution]"

If no source found after web search, say: "I could not find scholarly sources on this question." Do NOT say "scholars have not addressed this."

Rule 4: VALIDATE Mode Requires a Formal Verdict

When the user presents an analogy, illustration, or idea for validation, render one of four verdicts.

Format:

VERDICT: COMPATIBLE

The text does not specify the nature of Paul's thorn. Your analogy does not contradict 2 Cor 12:7-10. However:
- No lexical evidence supports "anxiety" specifically
- σκόλοψ = sharp stake/splinter; context favors physical affliction
- "Messenger of Satan" language has no modern psychological parallel

You may use this analogy pastorally, but present it as APPLICATION, not as what the text means.

Rule: Text evidence must be presented BEFORE the verdict. The verdict follows the evidence, not the other way around.

Wrong:

The analogy is pastorally useful as a point of contact but breaks down if pressed too hard exegetically.

(No verdict. Evidence not first. Drifted into pastoral application.)

Rule 5: No Devotional Drift

This skill produces scholarly analysis. Application is the user's job.

The skill explains what the text means. It does not:

• Tell the user what they should do
• Generate pastoral counseling
• Produce devotional content
• Tell the user how this "speaks to" their situation

Wrong:

God uses our anxiety for his purposes, and this passage reminds us to rest in his sufficient grace.

Correct:

The text presents divine power as perfected in human weakness (v.9b). The scope of application is the user's to determine.

Question Routing

The skill detects mode from the question. No user flag needed.

Mode 1: MEANING — triggered by:

• Questions about what a word, phrase, or sentence means
• "How would I explain X to a modern person?"
• Questions about grammatical function, voice, tense

Mode 2: VALIDATE — triggered by:

• User presents their own analogy, illustration, or comparison
• "Does this hold up?" / "Is this accurate?" / "Can I say..."
• User wants to check their idea against the text

Mode 3: CROSS-REFERENCE — triggered by:

• "What other passages connect to this?"
• Questions about thematic links, how a passage fits the biblical storyline
• "How does this relate to...?"

Ambiguous questions: Default to MEANING mode. If a question genuinely spans modes (e.g., "What does X mean and can I compare it to Y?"), handle both sequentially — MEANING first, then VALIDATE.

Topic mode (no passage): When no passage is provided, the skill enters topic mode. Warn immediately:

⚠️ TOPIC MODE: No specific passage provided. Confidence ceiling: MEDIUM.
MCP data cannot be verified without a passage. This answer draws on web search and training data.

Sub-Agent Delegation

This skill delegates data gathering and scholarly interpretation to the biblical-scholar agent, which internally delegates MCP data retrieval to the data-retriever agent (Haiku).

Delegation chain:

consult-biblical-scholar (skill, user's model)
  └─→ biblical-scholar (Sonnet) — scholarly analysis + source attribution
       └─→ data-retriever (Haiku) — MCP tool calls + compression

Mode mapping:

How to spawn:

Agent tool:
  subagent_type: "claude-of-alexandria:biblical-scholar"
  run_in_background: false
  prompt: "[MODE] [passage]. [question/claim]"

Every spawn in this chain is synchronous and unnamed. Both hops — this skill to biblical-scholar, and biblical-scholar to data-retriever — carry the same two requirements, and each exists to stop a specific silent failure.

What downstream consumers can rely on. Each agent in this chain is spawned through the ordinary subagent path, so the session's subagents/*.meta.json sidecar for every spawn carries a toolUseId linking it to the tool_use entry in its parent's transcript. Tools that reconstruct provenance by walking the subagent transcript family can assume that link exists for every agent this skill spawns, at every depth. The sidecar itself is written by the harness rather than by this plugin; what the plugin guarantees is that its spawns stay on the path where the harness writes a complete one.

A status line is not a result. If biblical-scholar returns a line saying its delegate is running or that analysis will follow, treat that as a failed delegation, not an answer. Fall back to direct MCP tool calls per the Rule 2 table and note the fallback. This is a distinct failure from the agent returning CANNOT ANSWER — a status line reports no confidence tier at all, so it cannot be inherited.

Parsing agent output:

• CONFIDENCE: line → inherit as skill's confidence ceiling
• VERDICT: line (VALIDATE mode) → use directly in skill output
• ## Scholarly Sources → integrate into Data Sources
• ## Limitations → surface in evidence summary

Fallback: If Agent tool fails, the agent returns CANNOT ANSWER, or the agent returns a spawn status instead of an analysis, fall back to direct MCP tool calls per Rule 2 table. Note the fallback in the response.

Workflow

1. Parse input
   → Extract passage (if any) and question
   → Detect mode: MEANING / VALIDATE / CROSS-REFERENCE
   → If no passage: enter topic mode, warn, cap confidence at MEDIUM

2. Delegate to biblical-scholar agent (if passage provided)
   → Spawn via Agent tool with mode-mapped prompt (see Sub-Agent Delegation)
   → The agent handles: pericope checking, MCP data gathering, scholarly interpretation
   → Parse agent response for: CONFIDENCE, analysis, verdict (if VALIDATE), sources
   → If agent spawn fails: fall back to direct MCP tool calls per Rule 2 table

2a. Call `query_controversies` for the passage and/or topic
   → Call BEFORE composing any prose response
   → Arg shape: `mode: "passage"` with `book` + `range` when a passage is identified;
     `mode: "topic"` with `topic` for open-ended questions (e.g., topic mode, Exodus dating);
     `mode: "list"` to enumerate all available controversy records.
     Example (passage): `query_controversies: {"mode": "passage", "book": "Exodus", "range": "12:1-40"}`
     Example (topic):   `query_controversies: {"mode": "topic", "topic": "exodus date"}`
   → If `topics` array is non-empty: insert the both-sides block per the Flagged Controversies rule above
   → If `topics` is empty: proceed without controversy block
   → Fallback: if the tool call fails, note "Controversy check unavailable" and proceed

3. Topic mode (if no passage)
   → Print the ⚠️ TOPIC MODE warning block FIRST, before any analysis.
     This warning is the FIRST text the user sees. Not after the answer.
     Not embedded in the confidence line. A standalone block at the top.
   → Web search to identify 2-3 key passages for the topic
   → Spawn biblical-scholar for at least ONE identified passage (mandatory — topic
     mode still requires MCP grounding on concrete text)
   → Confidence ceiling: MEDIUM

4. Web search (supplementary — only if agent's sources insufficient)
   → Search for scholarly commentary on passage + question topic
   → Evaluate source quality (Tier A/B/C/D)
   → For CROSS-REFERENCE: search specifically for intertextual connections
   → If search returns nothing useful: say so — do not fabricate consensus

5. State confidence tier
   → Inherit from biblical-scholar if delegation succeeded
   → Cap at MEDIUM if topic mode or if agent returned LOW/CANNOT ANSWER
   → Based only on what the agent and web search returned

6. Compose answer
   → Required: confidence declaration, evidence summary, answer, data sources
   → Mode-specific additions (see Output Format)
   → No devotional content, no unsourced claims

Output Format

Required in every response:

1. Topic mode warning (topic mode only) — the ⚠️ TOPIC MODE block is the FIRST thing output, before confidence, before anything else. Not optional.
2. Confidence tier — prominently (after topic mode warning if applicable)
3. Evidence summary — what MCP tools were called, what was found
4. Answer — scaled to complexity
5. Data sources — MCP tools queried + scholarly works cited

MEANING mode adds:

• Original language data with [query_morphology] attribution
• Discourse context (function in clause/paragraph)
• Modern explanation — labeled "for a contemporary audience," separated from technical data. Rule 5 still applies here. Explain what the Greek means in plain language. Do not shift into devotional register ("the promise here is…," "God doesn't start what he won't finish," "this reminds us…"). The section translates linguistics, not theology. Application remains the user's job.
• Scholarly positions if genuinely debated (with specific citations) — split confidence: morphology HIGH, debate MEDIUM. Do not resolve what scholars dispute.

VALIDATE mode adds:

• Text evidence first (MCP data on what the passage actually says)
• Then: VERDICT: SUPPORTED / COMPATIBLE / NOT SUPPORTED / INSUFFICIENT DATA
• Point-by-point comparison of analogy against text
• Usage guidance if SUPPORTED or COMPATIBLE

CROSS-REFERENCE mode adds:

• Ranked list of cross-references, each labeled:
  • Primary — shared lemma (query_vocabulary evidence) or explicit citation (query_ot_quotes)
  • Secondary — shared concept, different vocabulary (discourse/thematic evidence)
  • Entity — entity continuity (query_people, query_places, or query_person_network)
  • Editorial tradition — TSK/OpenBible-derived (query_cross_references). Always label as "editorial tradition candidate (N votes)" — NOT as Primary or Secondary evidence. These are tradition-based associations, not linguistically verified connections.
  • Scholarly — commentary-sourced (web search + citation)
• Each cross-reference includes WHY it's connected

Speaker attribution in MEANING/CROSS-REF modes:

• When analyzing dialogue passages, query_speakers data provides structured speaker attribution
• Divine speech markers are theologically significant for revelation claims
• Speaker transitions mark discourse boundaries
• Caveat: Angel-of-the-LORD attributions are dataset interpretations, not settled exegesis

Reference Data Access

NT Morphological Data

Call mcp__plugin_claude-of-alexandria_claude-of-alexandria-mcp__query_morphology with {"book": "[Book]", "range": "[chapter:verse-chapter:verse]"}

OT Morphological Data

Call mcp__plugin_claude-of-alexandria_claude-of-alexandria-mcp__query_morphology with {"book": "[Book]", "testament": "ot", "range": "[chapter:verse-chapter:verse]"}

Vocabulary and Lemma Frequencies

Call mcp__plugin_claude-of-alexandria_claude-of-alexandria-mcp__query_vocabulary with {"mode": "frequency", "book": "[Book]", "testament": "[nt|ot]"}

Levinsohn Discourse Features (NT)

Call mcp__plugin_claude-of-alexandria_claude-of-alexandria-mcp__query_discourse_features with {"book": "[Book]"}

Masoretic Paragraph Markers (OT)

Call mcp__plugin_claude-of-alexandria_claude-of-alexandria-mcp__query_paragraph_breaks with {"book": "[Book]"}

Red Flags

Theological Guardrails

Invocation

/consult-biblical-scholar Phil 1:6 What does "epiteleo" mean and how would I explain it to someone unfamiliar with Greek?
/consult-biblical-scholar 2 Cor 12:7-10 Can I compare Paul's thorn to chronic anxiety?
/consult-biblical-scholar Phil 2:5-11 Can I say Jesus couldn't help himself — his divine nature compelled him?
/consult-biblical-scholar Romans 3:25 What other passages connect to "hilasterion" and how?
/consult-biblical-scholar John 3:16 What does "monogenes" mean — only begotten or one and only?
/consult-biblical-scholar What is the biblical concept of Sabbath rest?

Output is inline (not saved to file). Every response includes confidence tier, MCP evidence summary, mode-appropriate answer, and data sources.