Passage glossary

Passage Glossary

Purpose

Produce a passage reader artifact: the passage text followed by a deduplicated, MCP-grounded glossary of every distinct headword occurring in it. Data-grounded. Never fabricate glosses. Output to file (default) or inline (--output print).

Key constraint: Every gloss must trace to MCP data (query_lexicon or morphology fallback). Unresolved headwords are shown explicitly — never invented.

---

Headword Identity

headword identity = strongs ?? lemma

A headword is identified by its Strong's number when the morphology data supplies one. When strongs is null, the exact lemma string serves as the key. This identity rule governs deduplication, lexicon calls, and glossary rendering throughout.

---

Pipeline (Five Steps — Execute in Order)

Step 1: Parse References

Accept one or more passage references. Supported forms:

• Single verse: John 1:1
• Intra-chapter range: John 1:1-5
• Cross-chapter range: John 1:1-2:11
• Multiple passages (comma- or +-separated): John 1:1-5, Romans 8:1-4

Translate each reference into the shared { book, range } shape used by query_morphology and bible_lookup. range format: "chapter:verse-chapter:verse" (e.g., "1:1-1:5").

One book per call. A cross-book span is split into per-book calls, each with its own query_morphology invocation.

Book not found: If query_morphology returns BOOK_NOT_FOUND, surface the error verbatim plus any suggestions it provides. Name the offending reference explicitly. Skip that passage but continue processing any remaining references — never silently drop a passage.

Do NOT call list_books — rely on the built-in error and suggestions from query_morphology.

---

Step 2: Extract and Deduplicate Headwords

For each { book, range }, call query_morphology with fields: "lexical". This returns per-word records containing lemma, strongs (nullable), and gloss (nullable).

Truncation check: After each call, inspect summary.truncated. If true, the range exceeded the server's 25k-character cap and trailing words were silently dropped. Split the range into smaller sub-ranges (e.g., by chapter) and re-query until summary.truncated is false for every sub-range. Never build a glossary from a truncated word list.

Deduplication: Across all words in all passages, apply the headword identity rule:

• Key = strongs when non-null
• Key = lemma string when strongs is null

Preserve first-occurrence order across the combined span. When two words share the same key, keep the first-seen record; discard the duplicate.

Each headword record carries:

• key: the dedup key (Strong's number or lemma string)
• lemma: the display form (first-seen surface form)
• strongs: the Strong's number, or null
• morphGloss: the morphology gloss field value (may be null)
• order: first-occurrence index (1-based)

When one Strong's number maps to multiple surface lemma forms in the passage, display the first-seen surface form and show the Strong's number on the glossary row for disambiguation.

---

Step 3: Resolve Glosses via Lexicon (Mandatory Batching ≤ 20)

Partition the distinct-headword list into two groups:

• Strongs group: headwords with a non-null strongs
• Null-strongs group: headwords whose key is the lemma string

Batching rule: Split each group into chunks of at most 20 entries per chunk. Do not exceed 20 per query_lexicon call — the API enforces a hard cap of 20 (strongs_ids or lemmas array max = 20). A passage with many distinct headwords requires multiple calls.

Example arithmetic: 47 distinct Strong's IDs → 3 calls: 20 + 20 + 7.

For the strongs group: call query_lexicon with mode: "strongs", strongs_ids: [<chunk of ≤20 Strong's numbers>].

For the null-strongs group: call query_lexicon with mode: "lemmas", lemmas: [<chunk of ≤20 lemma strings>].

Associating results:

• Lexicon returns entries keyed by strongs_id and not_found: string[] listing unresolved inputs.
• For the strongs group: match each entry by its strongs_id back to the headword's strongs key.
• For the null-strongs group: any input lemma absent from not_found was resolved; use the returned entry's gloss.

Gloss fallback chain (per headword, in order — never skip to a lower tier without exhausting the one above):

1. query_lexicon entry.gloss — authoritative, multi-source scholarly definition (Tier 1).
2. Morphology lexical gloss when it is non-null — labeled (morphology gloss) in output. Single-scholar provenance; use only when lexicon does not resolve the headword (Tier 2).
3. (unresolved — not in lexicon) — explicit label; no gloss invented (Tier 3 / anti-fabrication floor).

Transport failures vs. not_found: Distinguish:

• not_found: the headword is not in the lexicon — follow the fallback chain above.
• Transport error / API error on the call itself: retry the chunk once. If still failing after one retry, mark every headword in that chunk (gloss lookup failed) and continue. A complete headword list with some unresolved glosses is better than aborting the artifact.

Never fabricate a gloss. If no tier resolves a headword, output (unresolved — not in lexicon).

---

Step 4: Render the Artifact

Passage text: Call bible_lookup with { book, range } for each reference. Follow page.next_cursor until it is omitted, preserving the returned verse order. Surface any versification_note returned by the tool directly below the passage text.

Glossary: Project the resolved glosses back onto the ordered distinct-headword list from Step 2 (the authoritative, order-preserving spine). Do not render from lexicon merge order — use Step 2 order.

Per-row format (mandatory):

lemma [STRONG'S | (no-strongs)] — gloss

• Always show the Strong's number when present: e.g., θεός [G2316] — God
• When the headword had no strongs and was lemma-keyed, show the literal token (no-strongs): e.g., עֲבוֹדָ֔ה (no-strongs) — service
• Append transliteration when the lexicon returns it: e.g., θεός [G2316] theos — God
• Append (morphology gloss) label when using Tier 2 fallback.
• Append (unresolved — not in lexicon) or (gloss lookup failed) as applicable.

The per-row key token (STRONG'S number or (no-strongs)) MUST appear on every row — this is not optional decoration. It is the machine-readable dedup key that test assertions use to detect duplicate rows.

Section header: Use ## Glossary as the section heading.

Ordering: Default = first-occurrence (reading order). Alphabetical available on explicit user request.

Output modes:

• File mode (default, or --output file): Save to ~/.claude/passage-glossary/{book}/{YYYY-MM-DD}-{range}.md. Report the saved path.
• Print mode (--output print): Output the complete artifact inline. Do not save to file. Print ALL content — passage text plus full glossary — directly in the response.

Never ignore --output print. If the invocation includes --output print, print inline. Do not save to a file and return a summary.

Empty passage / no words: Emit the passage with an empty ## Glossary section and an explicit note: (No words found in this passage range.).

---

Step 5: Theological Guardrails

Every gloss in this artifact traces to MCP data (query_lexicon or morphology). Unresolved headwords are shown with explicit labels — they are never invented. This is the anti-fabrication floor: the artifact is a data-grounded study tool, not a commentary.

Do not add interpretive notes, theological commentary, or application content to glossary rows. The glossary lists headwords and their lexical definitions. Exegetical analysis belongs in /exegetical-notes, not here.

---

Common Failure Patterns (Red Flags)

Failure	Prevention
Glossary has duplicate headword rows	Apply dedup key = strongs ?? lemma before rendering; each key appears exactly once
Glossary omits words from the passage	Never render from a truncated morphology result; check summary.truncated and re-query
Wrong param name on morphology call	The param is fields: with value "lexical". There is no register parameter — do not use it.
Sent >20 IDs in one query_lexicon call	Batch into chunks of at most 20 — hard cap enforced by the API
Ignored not_found array	not_found is the unresolved signal; trigger the fallback chain on every entry in it
Fabricated a gloss when not_found	Follow the fallback chain; if Tier 2 (morphology gloss) is also null, output (unresolved — not in lexicon)
Rendered glossary in lexicon merge order	Render from the Step 2 ordered headword list, not from lexicon response iteration order
Per-row key token missing	Every row MUST show [G####] or (no-strongs) — it is the machine-readable dedup key
--output print but saved to file	If --output print is in the invocation, print ALL content inline

---

Invocation Format

/passage-glossary John 1:1-5
/passage-glossary John 1:1-5 --output print
/passage-glossary John 1:1-5, Romans 8:1-4
/passage-glossary Genesis 1:1-10

• --output: Optional. file (default) saves to disk. print outputs inline.
• Multiple passages: comma- or +-separated. Shared deduplication across all passages.
• Book names accept abbreviations (John, Gen, Rom, etc.) or full names.