@fortemi/core 2026.7.0 → 2026.7.2

package/dist/index.js

@@ -596,6 +596,19 @@ var migration0009 = {
   `
 };
 
+// src/migrations/0010_attachment_text_metadata.ts
+var migration0010 = {
+  version: 10,
+  name: "0010_attachment_text_metadata",
+  sql: `
+    ALTER TABLE attachment
+      ADD COLUMN IF NOT EXISTS mime_type TEXT;
+
+    ALTER TABLE attachment
+      ADD COLUMN IF NOT EXISTS extracted_text TEXT;
+  `
+};
+
 // src/migrations/index.ts
 var allMigrations = [
   migration0001,
@@ -606,7 +619,8 @@ var allMigrations = [
   migration0006,
   migration0007,
   migration0008,
-  migration0009
+  migration0009,
+  migration0010
 ];
 
 // src/data-archive.ts
@@ -1119,6 +1133,16 @@ function buildNoteConditions(options, startIdx, includeDeleted = false) {
 }
 
 // src/repositories/embedding-sets-repository.ts
+var ATTACHMENT_TEXT_JOIN = `
+       LEFT JOIN (
+         SELECT note_id,
+                string_agg(extracted_text, ' ' ORDER BY position, created_at)
+                  FILTER (WHERE extracted_text IS NOT NULL AND extracted_text <> '') as extracted_text
+         FROM attachment
+         WHERE deleted_at IS NULL
+         GROUP BY note_id
+       ) ax ON ax.note_id = n.id`;
+var COMBINED_TEXT_VECTOR_SQL = `to_tsvector('english', (coalesce(c.content, '') || ' ' || coalesce(ax.extracted_text, '')))`;
 var DEFAULT_COMPATIBILITY = {
   model: "require-same",
   dimension: "require-same",
@@ -1498,7 +1522,7 @@ var EmbeddingSetsRepository = class {
       params.push(criteria.updatedBefore);
     }
     if (criteria.query?.trim()) {
-      conditions.push(`(n.tsv @@ plainto_tsquery('english', $${idx}) OR to_tsvector('english', coalesce(c.content, '')) @@ plainto_tsquery('english', $${idx}))`);
+      conditions.push(`(n.tsv @@ plainto_tsquery('english', $${idx}) OR ${COMBINED_TEXT_VECTOR_SQL} @@ plainto_tsquery('english', $${idx}))`);
       params.push(criteria.query);
     }
     const result = await this.db.query(
@@ -1506,6 +1530,7 @@ var EmbeddingSetsRepository = class {
        FROM embedding e
        JOIN note n ON n.id = e.note_id
        LEFT JOIN note_revised_current c ON c.note_id = n.id
+       ${ATTACHMENT_TEXT_JOIN}
        WHERE ${conditions.join(" AND ")}
        ORDER BY e.note_id, e.created_at DESC`,
       params
@@ -1623,6 +1648,17 @@ var EmbeddingSetsRepository = class {
 };
 
 // src/repositories/search-repository.ts
+var ATTACHMENT_TEXT_JOIN2 = `
+       LEFT JOIN (
+         SELECT note_id,
+                string_agg(extracted_text, ' ' ORDER BY position, created_at)
+                  FILTER (WHERE extracted_text IS NOT NULL AND extracted_text <> '') as extracted_text
+         FROM attachment
+         WHERE deleted_at IS NULL
+         GROUP BY note_id
+       ) ax ON ax.note_id = n.id`;
+var COMBINED_TEXT_SQL = `(coalesce(c.content, '') || ' ' || coalesce(ax.extracted_text, ''))`;
+var COMBINED_TEXT_VECTOR_SQL2 = `to_tsvector('english', ${COMBINED_TEXT_SQL})`;
 var SearchRepository = class {
   constructor(db, semanticAvailable = false) {
     this.db = db;
@@ -1707,7 +1743,7 @@ var SearchRepository = class {
     const { conditions, params, nextIdx } = buildNoteConditions(options, 2);
     conditions.unshift(
       `(n.tsv @@ ${tsqFn}('english', $1) OR
-        to_tsvector('english', coalesce(c.content, '')) @@ ${tsqFn}('english', $1))`
+        ${COMBINED_TEXT_VECTOR_SQL2} @@ ${tsqFn}('english', $1))`
     );
     let paramIdx = this.scopeToResolvedEmbeddingSet(conditions, params, nextIdx, resolvedEmbeddingSet);
     const allParams = [query, ...params];
@@ -1716,6 +1752,7 @@ var SearchRepository = class {
       `SELECT COUNT(*) as count
        FROM note n
        LEFT JOIN note_revised_current c ON c.note_id = n.id
+       ${ATTACHMENT_TEXT_JOIN2}
        WHERE ${where}`,
       allParams
     );
@@ -1724,17 +1761,18 @@ var SearchRepository = class {
     const result = await this.db.query(
       `SELECT n.id, n.title, n.created_at, n.updated_at,
               ts_rank(
-                setweight(n.tsv, 'A') || setweight(to_tsvector('english', coalesce(c.content, '')), 'B'),
+                setweight(n.tsv, 'A') || setweight(${COMBINED_TEXT_VECTOR_SQL2}, 'B'),
                 ${tsqFn}('english', $1)
               ) as rank,
               ts_headline(
                 'english',
-                coalesce(c.content, ''),
+                ${COMBINED_TEXT_SQL},
                 ${tsqFn}('english', $1),
                 'StartSel=<mark>, StopSel=</mark>, MaxWords=35, MinWords=15'
               ) as snippet
        FROM note n
        LEFT JOIN note_revised_current c ON c.note_id = n.id
+       ${ATTACHMENT_TEXT_JOIN2}
        WHERE ${where}
        ORDER BY rank DESC, n.created_at DESC
        LIMIT $${paramIdx++} OFFSET $${paramIdx++}`,
@@ -1750,6 +1788,7 @@ var SearchRepository = class {
       const idsResult = await this.db.query(
         `SELECT n.id FROM note n
          LEFT JOIN note_revised_current c ON c.note_id = n.id
+         ${ATTACHMENT_TEXT_JOIN2}
          WHERE ${where}`,
         allParams
       );
@@ -1796,10 +1835,11 @@ var SearchRepository = class {
     const result = await this.db.query(
       `SELECT n.id, n.title, n.created_at, n.updated_at,
               (e.vector <=> $${vecIdx}::vector) as distance,
-              LEFT(coalesce(c.content, ''), 200) as snippet
+              LEFT(${COMBINED_TEXT_SQL}, 200) as snippet
        FROM embedding e
        JOIN note n ON n.id = e.note_id
        LEFT JOIN note_revised_current c ON c.note_id = n.id
+       ${ATTACHMENT_TEXT_JOIN2}
        WHERE ${where}
        ORDER BY e.vector <=> $${vecIdx}::vector ASC
        LIMIT $${limIdx} OFFSET $${offIdx}`,
@@ -1840,7 +1880,7 @@ var SearchRepository = class {
     const textConditions = [
       ...textCond.conditions,
       `(n.tsv @@ ${tsqFn}('english', $1) OR
-        to_tsvector('english', coalesce(c.content, '')) @@ ${tsqFn}('english', $1))`
+        ${COMBINED_TEXT_VECTOR_SQL2} @@ ${tsqFn}('english', $1))`
     ];
     this.scopeToResolvedEmbeddingSet(textConditions, textCond.params, textCond.nextIdx, resolvedEmbeddingSet);
     const textWhere = textConditions.join(" AND ");
@@ -1848,11 +1888,12 @@ var SearchRepository = class {
     const textResult = await this.db.query(
       `SELECT n.id,
               ts_rank(
-                setweight(n.tsv, 'A') || setweight(to_tsvector('english', coalesce(c.content, '')), 'B'),
+                setweight(n.tsv, 'A') || setweight(${COMBINED_TEXT_VECTOR_SQL2}, 'B'),
                 ${tsqFn}('english', $1)
               ) as rank
        FROM note n
        LEFT JOIN note_revised_current c ON c.note_id = n.id
+       ${ATTACHMENT_TEXT_JOIN2}
        WHERE ${textWhere}
        ORDER BY rank DESC
        LIMIT 100`,
@@ -1886,9 +1927,10 @@ var SearchRepository = class {
     }
     const noteResult = await this.db.query(
       `SELECT n.id, n.title, n.created_at, n.updated_at,
-              LEFT(coalesce(c.content, ''), 200) as snippet
+              LEFT(${COMBINED_TEXT_SQL}, 200) as snippet
        FROM note n
        LEFT JOIN note_revised_current c ON c.note_id = n.id
+       ${ATTACHMENT_TEXT_JOIN2}
        WHERE n.id = ANY($1)`,
       [pageIds]
     );
@@ -1936,9 +1978,10 @@ var SearchRepository = class {
     const listParams = [...params, limit, offset];
     const result = await this.db.query(
       `SELECT n.id, n.title, n.created_at, n.updated_at,
-              LEFT(coalesce(c.content, ''), 200) as snippet
+              LEFT(${COMBINED_TEXT_SQL}, 200) as snippet
        FROM note n
        LEFT JOIN note_revised_current c ON c.note_id = n.id
+       ${ATTACHMENT_TEXT_JOIN2}
        WHERE ${where}
        ORDER BY n.created_at DESC
        LIMIT $${paramIdx++} OFFSET $${paramIdx++}`,
@@ -3811,6 +3854,29 @@ function getLlmFunction() {
   return llmFn;
 }
 
+// src/repositories/note-text.ts
+async function getNoteTextWithExtractedAttachments(db, noteId) {
+  const result = await db.query(
+    `SELECT c.content,
+            COALESCE(string_agg(a.extracted_text, E'
+' ORDER BY a.position, a.created_at)
+              FILTER (WHERE a.extracted_text IS NOT NULL AND a.extracted_text <> ''), '') as extracted_text
+       FROM note_revised_current c
+       LEFT JOIN attachment a ON a.note_id = c.note_id AND a.deleted_at IS NULL
+       WHERE c.note_id = $1
+       GROUP BY c.content`,
+    [noteId]
+  );
+  if (result.rows.length === 0) return null;
+  const content = result.rows[0].content;
+  const extractedText = result.rows[0].extracted_text ?? "";
+  return {
+    content,
+    extracted_text: extractedText,
+    combined: [content, extractedText].filter(Boolean).join("\n")
+  };
+}
+
 // src/job-queue-worker.ts
 var JOB_PRIORITIES = {
   ai_revision: 1,
@@ -4092,12 +4158,9 @@ function conceptTaggingHandler(job, db) {
   return (async () => {
     const llmFn2 = getLlmFunction();
     if (!llmFn2) return { skipped: true, reason: "no LLM function registered" };
-    const result = await db.query(
-      `SELECT content FROM note_revised_current WHERE note_id = $1`,
-      [job.note_id]
-    );
-    if (result.rows.length === 0) throw new Error(`No content found for note ${job.note_id}`);
-    const content = result.rows[0].content;
+    const noteText = await getNoteTextWithExtractedAttachments(db, job.note_id);
+    if (!noteText) throw new Error(`No content found for note ${job.note_id}`);
+    const content = noteText.combined;
     const prompt = `Task: Extract 3-5 topic tags from the text below.
 Rules:
 - Each tag is 1-3 words, lowercase
@@ -4990,9 +5053,17 @@ var AttachmentsRepository = class {
     }
     const attachmentId = generateId();
     await this.db.query(
-      `INSERT INTO attachment (id, note_id, blob_id, filename, display_name)
-       VALUES ($1, $2, $3, $4, $5)`,
-      [attachmentId, input.noteId, blobId, input.filename, input.displayName ?? null]
+      `INSERT INTO attachment (id, note_id, blob_id, filename, display_name, mime_type, extracted_text)
+       VALUES ($1, $2, $3, $4, $5, $6, $7)`,
+      [
+        attachmentId,
+        input.noteId,
+        blobId,
+        input.filename,
+        input.displayName ?? null,
+        input.mimeType ?? null,
+        input.extractedText ?? null
+      ]
     );
     return this.get(attachmentId);
   }
@@ -5055,6 +5126,7 @@ var ManageAttachmentsInputSchema = z.object({
   data_base64: z.string().optional(),
   filename: z.string().optional(),
   mime_type: z.string().optional(),
+  extracted_text: z.string().optional(),
   display_name: z.string().optional()
 });
 async function manageAttachments(db, blobStore, rawInput) {
@@ -5073,6 +5145,7 @@ async function manageAttachments(db, blobStore, rawInput) {
         data,
         filename: input.filename,
         mimeType: input.mime_type,
+        extractedText: input.extracted_text,
         displayName: input.display_name
       });
       return { action: "attach", attachment, size_bytes: data.length };
@@ -5324,12 +5397,9 @@ function averageEmbeddings(embeddings) {
 async function embeddingGenerationHandler(job, db) {
   const fn = embedFn;
   if (!fn) return { skipped: true, reason: "no embed function registered" };
-  const noteResult = await db.query(
-    `SELECT content FROM note_revised_current WHERE note_id = $1`,
-    [job.note_id]
-  );
-  if (noteResult.rows.length === 0) throw new Error(`No content for note ${job.note_id}`);
-  const content = noteResult.rows[0].content;
+  const noteText = await getNoteTextWithExtractedAttachments(db, job.note_id);
+  if (!noteText) throw new Error(`No content for note ${job.note_id}`);
+  const content = noteText.combined;
   const chunks = chunkText(content);
   const embeddings = await fn(chunks);
   const vector2 = averageEmbeddings(embeddings);
@@ -6512,6 +6582,7 @@ function noteToShard(note) {
     title: note.title,
     original_content: note.original_content,
     revised_content: note.revised_content,
+    ...note.binary_sources?.length ? { binary_sources: note.binary_sources } : {},
     format: note.format,
     source: note.source,
     starred: note.is_starred,
@@ -6532,6 +6603,7 @@ function noteFromShard(shard) {
     is_archived: shard.archived,
     original_content: shard.original_content,
     revised_content: shard.revised_content,
+    binary_sources: shard.binary_sources,
     tags: shard.tags,
     created_at: shard.created_at,
     updated_at: shard.updated_at,
@@ -6789,9 +6861,40 @@ async function exportShard(db, options) {
     tags.push(row.tag);
     tagsByNote.set(row.note_id, tags);
   }
+  const attachmentRows = await db.query(
+    `SELECT a.note_id,
+            a.id,
+            a.filename,
+            a.mime_type,
+            a.extracted_text,
+            b.content_hash,
+            b.size_bytes,
+            b.storage_path
+       FROM attachment a
+       JOIN attachment_blob b ON b.id = a.blob_id
+       WHERE a.deleted_at IS NULL
+       ORDER BY a.note_id, a.position, a.created_at`
+  );
+  const binarySourcesByNote = /* @__PURE__ */ new Map();
+  for (const row of attachmentRows.rows) {
+    const source = {
+      extracted_text: row.extracted_text ?? "",
+      attachment: {
+        id: row.id,
+        path: row.storage_path ?? row.filename,
+        mime: row.mime_type,
+        checksum: row.content_hash,
+        bytes: Number(row.size_bytes)
+      }
+    };
+    const sources = binarySourcesByNote.get(row.note_id) ?? [];
+    sources.push(source);
+    binarySourcesByNote.set(row.note_id, sources);
+  }
   const notes = noteRows.rows.map((row) => ({
     ...row,
-    tags: tagsByNote.get(row.id) ?? []
+    tags: tagsByNote.get(row.id) ?? [],
+    binary_sources: binarySourcesByNote.get(row.id)
   }));
   const exportedNoteIds = new Set(notes.map((n) => n.id));
   const shardNotes = notes.map((n) => noteToShard(n));
@@ -7653,7 +7756,8 @@ function tokenize(query) {
   return query.toLowerCase().split(/[^a-z0-9]+/i).filter((token) => token.length > 0);
 }
 function noteSearchText(note) {
-  return `${note.title ?? ""} ${note.original_content} ${note.revised_content ?? ""}`.toLowerCase();
+  const extractedText = note.binary_sources?.map((source) => source.extracted_text).filter(Boolean).join(" ") ?? "";
+  return `${note.title ?? ""} ${note.original_content} ${note.revised_content ?? ""} ${extractedText}`.toLowerCase();
 }
 function countOccurrences(haystack, needle) {
   if (!needle) return 0;
@@ -7674,7 +7778,8 @@ function noteMatchesTokens(note, tokens) {
 function rankNote(note, tokens, weights) {
   if (tokens.length === 0) return 0;
   const title = (note.title ?? "").toLowerCase();
-  const content = `${note.original_content} ${note.revised_content ?? ""}`.toLowerCase();
+  const extractedText = note.binary_sources?.map((source) => source.extracted_text).filter(Boolean).join(" ") ?? "";
+  const content = `${note.original_content} ${note.revised_content ?? ""} ${extractedText}`.toLowerCase();
   const tagText = note.tags.join(" ").toLowerCase();
   let score = 0;
   for (const token of tokens) {
@@ -7685,7 +7790,8 @@ function rankNote(note, tokens, weights) {
   return score;
 }
 function makeSnippet(note, tokens, length) {
-  const content = `${note.original_content} ${note.revised_content ?? ""}`.trim();
+  const extractedText = note.binary_sources?.map((source) => source.extracted_text).filter(Boolean).join(" ") ?? "";
+  const content = `${note.original_content} ${note.revised_content ?? ""} ${extractedText}`.trim();
   if (tokens.length === 0) return content.slice(0, length);
   const lower = content.toLowerCase();
   const firstAt = tokens.map((token) => lower.indexOf(token)).filter((index) => index !== -1).sort((a, b) => a - b)[0] ?? 0;
@@ -8093,8 +8199,6 @@ var REQUIRED_RECORD_FIELDS = [
   "id",
   "type",
   "source",
-  "title",
-  "text",
   "facets",
   "tags",
   "concepts",
@@ -8135,6 +8239,12 @@ function isPlainRecord(value) {
 function isOptionalStringArray(value) {
   return value === void 0 || Array.isArray(value) && value.every((item) => typeof item === "string");
 }
+function isSupportedIndexSchemaVersion(value) {
+  return value === "aiwg.fortemi.index.export.v1" || value === "aiwg.fortemi.index.export.v2";
+}
+function isSupportedRecordSchemaVersion(value) {
+  return value === "aiwg.fortemi.index.record.v1" || value === "aiwg.fortemi.index.record.v2";
+}
 function validateOptionalRichMetadata(item, index, errors) {
   if (item.skos_concepts !== void 0) {
     if (!Array.isArray(item.skos_concepts)) {
@@ -8181,28 +8291,80 @@ function validateOptionalRichMetadata(item, index, errors) {
       if (relationship.metadata !== void 0 && !isPlainRecord(relationship.metadata)) {
         errors.push("items[" + index + "].relationships[" + relationshipIndex + "].metadata must be an object");
       }
+      if (relationship.direction !== void 0 && relationship.direction !== "upstream" && relationship.direction !== "downstream" && relationship.direction !== "related") {
+        errors.push("items[" + index + "].relationships[" + relationshipIndex + "].direction must be upstream, downstream, or related");
+      }
+      if (relationship.target_path !== void 0 && typeof relationship.target_path !== "string") {
+        errors.push("items[" + index + "].relationships[" + relationshipIndex + "].target_path must be a string");
+      }
     }
   }
+  if (item.search !== void 0) {
+    if (!isPlainRecord(item.search)) {
+      errors.push("items[" + index + "].search must be an object");
+    } else {
+      if (!isOptionalStringArray(item.search.triggers)) errors.push("items[" + index + "].search.triggers must be a string array");
+      if (!isOptionalStringArray(item.search.aliases)) errors.push("items[" + index + "].search.aliases must be a string array");
+      if (!isOptionalStringArray(item.search.tags)) errors.push("items[" + index + "].search.tags must be a string array");
+      if (item.search.frontmatter !== void 0 && !isPlainRecord(item.search.frontmatter)) {
+        errors.push("items[" + index + "].search.frontmatter must be an object");
+      }
+    }
+  }
+  if (item.chunks !== void 0) {
+    if (!Array.isArray(item.chunks)) {
+      errors.push("items[" + index + "].chunks must be an array when present");
+    } else {
+      for (const [chunkIndex, chunk] of item.chunks.entries()) {
+        if (!isPlainRecord(chunk)) errors.push("items[" + index + "].chunks[" + chunkIndex + "] must be an object");
+        if (chunk.metadata !== void 0 && !isPlainRecord(chunk.metadata)) {
+          errors.push("items[" + index + "].chunks[" + chunkIndex + "].metadata must be an object");
+        }
+      }
+    }
+  }
+  if (item.embeddings !== void 0) {
+    if (!Array.isArray(item.embeddings)) {
+      errors.push("items[" + index + "].embeddings must be an array when present");
+    } else {
+      for (const [embeddingIndex, embedding] of item.embeddings.entries()) {
+        if (!isPlainRecord(embedding)) errors.push("items[" + index + "].embeddings[" + embeddingIndex + "] must be an object");
+        const vector2 = embedding.embedding ?? embedding.vector;
+        if (vector2 !== void 0 && (!Array.isArray(vector2) || !vector2.every((entry) => typeof entry === "number"))) {
+          errors.push("items[" + index + "].embeddings[" + embeddingIndex + "].embedding/vector must be a number array");
+        }
+        if (embedding.metadata !== void 0 && !isPlainRecord(embedding.metadata)) {
+          errors.push("items[" + index + "].embeddings[" + embeddingIndex + "].metadata must be an object");
+        }
+      }
+    }
+  }
+  if (item.compatibility !== void 0 && !isPlainRecord(item.compatibility)) {
+    errors.push("items[" + index + "].compatibility must be an object");
+  }
 }
 function validateAiwgFortemiIndexExport(value) {
   const errors = [];
   const counts = {};
   const data = value;
-  if (data?.schema_version !== "aiwg.fortemi.index.export.v1") {
-    errors.push("schema_version must be aiwg.fortemi.index.export.v1");
+  if (!isSupportedIndexSchemaVersion(data?.schema_version)) {
+    errors.push("schema_version must be aiwg.fortemi.index.export.v1 or aiwg.fortemi.index.export.v2");
   }
   if (!hasString(data?.generated_at)) errors.push("generated_at is required");
   if (!hasString(data?.source?.repo)) errors.push("source.repo is required");
   if (!hasString(data?.source?.privacy)) errors.push("source.privacy is required");
   if (!Array.isArray(data?.items)) errors.push("items must be an array");
+  if (data.compatibility !== void 0 && !isPlainRecord(data.compatibility)) {
+    errors.push("compatibility must be an object");
+  }
   const ids = /* @__PURE__ */ new Set();
   let previousId = "";
   for (const [index, item] of (data.items ?? []).entries()) {
     for (const field of REQUIRED_RECORD_FIELDS) {
       if (!(field in item)) errors.push("items[" + index + "]." + field + " is required");
     }
-    if (item.schema_version !== "aiwg.fortemi.index.record.v1") {
-      errors.push("items[" + index + "].schema_version must be aiwg.fortemi.index.record.v1");
+    if (!isSupportedRecordSchemaVersion(item.schema_version)) {
+      errors.push("items[" + index + "].schema_version must be aiwg.fortemi.index.record.v1 or aiwg.fortemi.index.record.v2");
     }
     if (!hasString(item.id)) errors.push("items[" + index + "].id is required");
     if (hasString(item.id) && ids.has(item.id)) errors.push("duplicate id: " + item.id);
@@ -8216,6 +8378,12 @@ function validateAiwgFortemiIndexExport(value) {
     if (!hasString(item.source?.path)) errors.push("items[" + index + "].source.path is required");
     if (!hasString(item.source?.repo_relative_path)) errors.push("items[" + index + "].source.repo_relative_path is required");
     if (!hasString(item.source?.locator)) errors.push("items[" + index + "].source.locator is required");
+    if (typeof item.title !== "string" && typeof item.search?.title !== "string" && typeof item.search?.name !== "string") {
+      errors.push("items[" + index + "].title or search.title/search.name is required");
+    }
+    if (typeof item.text !== "string" && typeof item.search?.body !== "string" && typeof item.search?.summary !== "string") {
+      errors.push("items[" + index + "].text or search.body/search.summary is required");
+    }
     if (!Array.isArray(item.tags)) errors.push("items[" + index + "].tags must be an array");
     if (!Array.isArray(item.concepts)) errors.push("items[" + index + "].concepts must be an array");
     if (!Array.isArray(item.relationships)) errors.push("items[" + index + "].relationships must be an array");
@@ -8296,8 +8464,8 @@ function validateProjectedRecords(items) {
   const ids = /* @__PURE__ */ new Set();
   let previousId = "";
   for (const [index, item] of items.entries()) {
-    if (item.schema_version !== "aiwg.fortemi.index.record.v1") {
-      errors.push("items[" + index + "].schema_version must be aiwg.fortemi.index.record.v1");
+    if (!isSupportedRecordSchemaVersion(item.schema_version)) {
+      errors.push("items[" + index + "].schema_version must be aiwg.fortemi.index.record.v1 or aiwg.fortemi.index.record.v2");
     }
     if (!hasString(item.id)) errors.push("items[" + index + "].id is required");
     if (hasString(item.id) && ids.has(item.id)) errors.push("duplicate id: " + item.id);
@@ -8307,8 +8475,12 @@ function validateProjectedRecords(items) {
     }
     if (hasString(item.id)) previousId = item.id;
     if (!hasString(item.type)) errors.push("items[" + index + "].type must be a non-empty string");
-    if (!hasString(item.title)) errors.push("items[" + index + "].title is required");
-    if (typeof item.text !== "string") errors.push("items[" + index + "].text is required");
+    if (typeof item.title !== "string" && typeof item.search?.title !== "string" && typeof item.search?.name !== "string") {
+      errors.push("items[" + index + "].title or search.title/search.name is required");
+    }
+    if (typeof item.text !== "string" && typeof item.search?.body !== "string" && typeof item.search?.summary !== "string") {
+      errors.push("items[" + index + "].text or search.body/search.summary is required");
+    }
     if (!item.facets || typeof item.facets !== "object" || Array.isArray(item.facets)) {
       errors.push("items[" + index + "].facets must be an object");
     }
@@ -8400,6 +8572,86 @@ function getAiwgFortemiFacets(items) {
   }
   return result;
 }
+function recordTitle(item) {
+  return item.title ?? item.search?.title ?? item.search?.name ?? item.id;
+}
+function recordText(item) {
+  const base = item.text ?? item.search?.body ?? item.search?.summary ?? item.chunks?.map((chunk) => chunk.text ?? chunk.body ?? chunk.summary ?? "").filter(Boolean).join("\n") ?? "";
+  const extractedText = "binary_sources" in item ? item.binary_sources?.map((source) => source.extracted_text).filter(Boolean).join("\n") ?? "" : "";
+  return [base, extractedText].filter(Boolean).join("\n");
+}
+function defaultEmbeddingInput(record, granularity) {
+  const title = recordTitle(record);
+  const text = recordText(record);
+  if (granularity === "title-summary") {
+    const extractedText = record.binary_sources?.map((source) => source.extracted_text).filter(Boolean).join("\n") ?? "";
+    return [title, record.search?.summary ?? "", extractedText].filter(Boolean).join("\n");
+  }
+  return [title, text].filter(Boolean).join("\n");
+}
+function generatedAtString(value) {
+  if (value instanceof Date) return value.toISOString();
+  return value ?? (/* @__PURE__ */ new Date()).toISOString();
+}
+async function buildAiwgStaticEmbeddingSet(index, options) {
+  assertAiwgFortemiIndexExport(index);
+  const granularity = options.granularity ?? "body";
+  const records = options.records ?? index.items;
+  const embeddings = [];
+  for (const record of records) {
+    const input = options.textForRecord?.(record) ?? defaultEmbeddingInput(record, granularity);
+    const embedding = await options.backend.embed(input, record);
+    if (embedding.length !== options.backend.dimensions) {
+      throw new Error(`Embedding for ${record.id} has ${embedding.length} dimensions; expected ${options.backend.dimensions}`);
+    }
+    embeddings.push({
+      record_id: record.id,
+      embedding,
+      granularity,
+      input_hash: computeHash(new TextEncoder().encode(input)),
+      source_path: record.source.path
+    });
+  }
+  const embeddingSet = {
+    schema_version: "aiwg.fortemi.embedding.set.v1",
+    id: options.id,
+    model: options.backend.model,
+    dimensions: options.backend.dimensions,
+    generated_at: generatedAtString(options.generatedAt),
+    granularity,
+    ...options.metric ? { metric: options.metric } : {},
+    input_hash_algorithm: "sha256",
+    embeddings
+  };
+  assertAiwgStaticEmbeddingSet(embeddingSet);
+  return embeddingSet;
+}
+function recordSearchValues(item) {
+  const search = item.search;
+  const values = [
+    item.id,
+    recordTitle(item),
+    recordText(item),
+    search?.title,
+    search?.name,
+    search?.summary,
+    search?.body,
+    search?.capability,
+    search?.phase,
+    search?.type,
+    ...search?.triggers ?? [],
+    ...search?.aliases ?? [],
+    ...search?.tags ?? [],
+    ...(item.chunks ?? []).flatMap((chunk) => [chunk.text, chunk.body, chunk.summary, chunk.source_path])
+  ];
+  if (search?.frontmatter) {
+    for (const value of Object.values(search.frontmatter)) {
+      if (typeof value === "string") values.push(value);
+      else if (Array.isArray(value)) values.push(...value.filter((entry) => typeof entry === "string"));
+    }
+  }
+  return values.filter((value) => typeof value === "string" && value.length > 0);
+}
 function buildAiwgChunkedIndex(index, options = {}) {
   const partSize = hasPositiveInteger(options.partSize) ? options.partSize : 500;
   const projection = options.projection;
@@ -8461,14 +8713,21 @@ function matchesFacetFilters(item, filters) {
 function queryMatches(item, q) {
   if (!q) return [];
   const matches = [];
-  if (item.title.toLowerCase().includes(q)) matches.push({ field: "title", value: item.title });
-  if (item.text.toLowerCase().includes(q)) matches.push({ field: "text", value: item.text });
+  const title = recordTitle(item);
+  const text = recordText(item);
+  if (title.toLowerCase().includes(q)) matches.push({ field: "title", value: title });
+  if (text.toLowerCase().includes(q)) matches.push({ field: "text", value: text });
   for (const tag of item.tags) {
     if (tag.toLowerCase().includes(q)) matches.push({ field: "tag", value: tag });
   }
   for (const concept of item.concepts) {
     if (concept.toLowerCase().includes(q)) matches.push({ field: "concept", value: concept });
   }
+  for (const value of recordSearchValues(item)) {
+    if (value !== title && value !== text && value.toLowerCase().includes(q)) {
+      matches.push({ field: "text", value, score: DEFAULT_QUERY_WEIGHTS.text });
+    }
+  }
   return matches;
 }
 var DISCOVERY_STOPWORDS = /* @__PURE__ */ new Set([
@@ -8524,17 +8783,28 @@ function discoveryMatches(item, query) {
   const idParts = item.id.split(/[:/]/);
   const names = [
     item.id,
-    item.title,
+    recordTitle(item),
+    item.search?.name,
+    item.search?.title,
+    ...item.search?.aliases ?? [],
     ...idParts,
     ...facetValues(item, ["name", "canonical_name", "command", "skill", "agent", "rule"])
-  ].filter(Boolean);
-  const triggers = facetValues(item, ["trigger", "triggers", "trigger_phrase", "trigger_phrases"]);
+  ].filter((value) => hasString(value));
+  const triggers = [
+    ...facetValues(item, ["trigger", "triggers", "trigger_phrase", "trigger_phrases"]),
+    ...item.search?.triggers ?? []
+  ];
   const capabilities = [
     ...facetValues(item, ["capability", "capabilities", "summary", "description"]),
+    item.search?.capability,
+    item.search?.summary,
+    item.search?.phase,
+    item.search?.type,
+    ...item.search?.tags ?? [],
     ...item.concepts,
     ...item.tags
-  ];
-  const sourceValues = [item.source?.path, item.source?.repo_relative_path, item.source?.locator].filter(Boolean);
+  ].filter((value) => hasString(value));
+  const sourceValues = [item.source?.path, item.source?.repo_relative_path, item.source?.locator].filter((value) => hasString(value));
   for (const name of names) {
     const canonicalName = canonicalDiscoveryName(name);
     if (!canonicalName) continue;
@@ -8544,8 +8814,9 @@ function discoveryMatches(item, query) {
       addDiscoveryMatch(matches, { field: "id", value: name, score: 48, reason: "near canonical name" });
     }
   }
-  const titleOverlap = tokenOverlapScore(tokens, item.title);
-  if (titleOverlap > 0) addDiscoveryMatch(matches, { field: "title", value: item.title, score: 18 * titleOverlap, reason: "title token overlap" });
+  const title = recordTitle(item);
+  const titleOverlap = tokenOverlapScore(tokens, title);
+  if (titleOverlap > 0) addDiscoveryMatch(matches, { field: "title", value: title, score: 18 * titleOverlap, reason: "title token overlap" });
   for (const trigger of triggers) {
     const overlap = tokenOverlapScore(tokens, trigger);
     if (overlap > 0) addDiscoveryMatch(matches, { field: "facet", value: trigger, score: 34 * overlap, reason: "trigger phrase" });
@@ -8554,8 +8825,9 @@ function discoveryMatches(item, query) {
     const overlap = tokenOverlapScore(tokens, capability);
     if (overlap > 0) addDiscoveryMatch(matches, { field: "concept", value: capability, score: 22 * overlap, reason: "capability overlap" });
   }
-  const textOverlap = tokenOverlapScore(tokens, item.text);
-  if (textOverlap > 0) addDiscoveryMatch(matches, { field: "text", value: item.text, score: 8 * textOverlap, reason: "body token overlap" });
+  const text = recordText(item);
+  const textOverlap = tokenOverlapScore(tokens, text);
+  if (textOverlap > 0) addDiscoveryMatch(matches, { field: "text", value: text, score: 8 * textOverlap, reason: "body token overlap" });
   for (const source of sourceValues) {
     const overlap = tokenOverlapScore(tokens, source);
     if (overlap > 0) addDiscoveryMatch(matches, { field: "source", value: source, score: 2 * overlap, reason: "path overlap" });
@@ -8583,7 +8855,7 @@ function createSnippet(item, matches, q, maxLength) {
   const textMatch = matches.find((match) => match.field === "text");
   const titleMatch = matches.find((match) => match.field === "title");
   const firstMatch = textMatch ?? titleMatch ?? matches[0];
-  return clipSnippet(firstMatch?.value ?? item.text, q, maxLength);
+  return clipSnippet(firstMatch?.value ?? recordText(item), q, maxLength);
 }
 function createRankedEntries(items, q, options, ordinalBase = 0) {
   const weights = { ...DEFAULT_QUERY_WEIGHTS, ...options.weights };
@@ -8845,13 +9117,16 @@ function edgeFromRelationship(sourceId, relationship) {
     source_id: sourceId,
     target_id: relationship.target_id,
     type: relationship.type,
-    ...relationship.source_path ? { source_path: relationship.source_path } : {}
+    ...relationship.source_path ? { source_path: relationship.source_path } : {},
+    ...relationship.target_path ? { target_path: relationship.target_path } : {},
+    ...relationship.direction ? { direction: relationship.direction } : {}
   };
 }
 function relationshipMatches(edge, options = {}) {
   const type = relationshipTypeFilter(options);
   const direction = options.direction ?? "both";
   if (type && edge.type !== type) return false;
+  if (options.relationshipDirection && edge.direction !== options.relationshipDirection) return false;
   if (options.sourceId && edge.source_id !== options.sourceId) return false;
   if (options.targetId && edge.target_id !== options.targetId) return false;
   if (direction === "out" && options.targetId && edge.target_id !== options.targetId) return false;
@@ -8859,7 +9134,7 @@ function relationshipMatches(edge, options = {}) {
   return true;
 }
 function nodeSummary(item) {
-  return { id: item.id, type: item.type, title: item.title };
+  return { id: item.id, type: item.type, title: recordTitle(item) };
 }
 function addNode(nodes, item) {
   if (item) nodes.set(item.id, nodeSummary(item));
@@ -9243,8 +9518,8 @@ function communityIdsFor(item, options) {
 }
 
 // src/index.ts
-var VERSION = "2026.7.0";
+var VERSION = "2026.7.2";
 
-export { AIWG_SCAN_REQUIRED_FIELDS, ArchiveManager, AttachmentsRepository, CURRENT_MIGRATION_HEAD, CURRENT_SHARD_VERSION, CapabilityManager, CaptureKnowledgeInputSchema, CollectionsRepository, CommunitiesRepository, DB_SNAPSHOT_SCHEMA_VERSION, DbSnapshotVersionError, EMBED_REQUEST_KIND, EMBED_RESPONSE_KIND, EmbeddingSetsRepository, FallbackRouter, FortemiToolManifest, GetNoteInputSchema, GraphRepository, JOB_CAPABILITIES, JOB_PRIORITIES, JobQueueWorker, LOCAL_ENDPOINTS, LinksRepository, ListNotesInputSchema, ManageArchiveInputSchema, ManageAttachmentsInputSchema, ManageCapabilitiesInputSchema, ManageCollectionsInputSchema, ManageLinksInputSchema, ManageNoteInputSchema, ManageTagsInputSchema, MemoryBlobStore, MigrationRunner, NotesRepository, OpenAICompatibleProvider, PGliteStorageBackend, PGliteStorageBackendFactory, PGliteWorkerClient, PGliteWorkerStorageBackend, PGliteWorkerStorageBackendFactory, ProvenanceRepository, ProviderRegistry, SHARD_FORMAT, SUPPORTED_PGLITE_VERSION, SearchInputSchema, SearchRepository, SkosRepository, TagsRepository, TransactionProxy, TypedEventBus, VERSION, aiRevisionHandler, aiwgFortemiIndexToCommunityGraph, allMigrations, appendPluginScript, assertAiwgFortemiChunkManifest, assertAiwgFortemiChunkPart, assertAiwgFortemiIndexExport, assertAiwgStaticEmbeddingSet, buildAiwgChunkedIndex, buildNoteConditions, buildPluginCsp, captureKnowledge, chunkText, classifyError, classifyModel, clearPrefetchedShard, collectionFromShard, collectionToShard, computeHash, computeSri, conceptTaggingHandler, cosineSimilarity, createAiwgFetchChunkLoader, createAiwgFetchDetailLoader, createAiwgIndexController, createAiwgReviewDecisionExport, createBlobStore, createCosineSemanticProvider, createCspReportHandler, createFortemi, createLegacyProvider, createPGliteBackend, createPGliteInstance, createRemoteBackend, createRoutes, createShardBackend, createWorkerEmbedFunction, defaultStorageBackendFactory, detectCommunities, detectGpuCapabilities, detectInferenceCapabilities, discoverLocalProviders, dumpDbSnapshot, embeddingFromShard, embeddingGenerationHandler, embeddingSetFromShard, embeddingSetMemberToShard, embeddingSetToShard, embeddingToShard, enqueueFullWorkflow, enqueueJob, enqueueNoteCreationJobs, estimateModelFit, estimateVramMB, estimateVramTier, exportShard, fetchPluginScript, findAiwgStaticDuplicatePairs, fortemiManifest, fromPrefetched, generateId, getAiwgFortemiFacets, getEmbedFunction, getFortemiBridge, getFortemiSecretStore, getJobQueueStatus, getLlmFunction, getNote, getPrefetchedSha256, handleEmbedRequests, hasFortemiSecureSecrets, importShard, isPluginScriptAllowed, isShardPrefetched, linkFromShard, linkToShard, linkingHandler, listNotes, manageArchive, manageAttachments, manageCapabilities, manageCollections, manageLinks, manageNote, manageTags, matchRoute, noteFromShard, noteSkosTagToShard, noteToShard, openShard, packTarGz, parseCspReport, prefetchShard, provenanceEdgeToShard, queryAiwgFortemiIndex, queryAiwgHybridIndex, queryAiwgSemanticIndex, registerLlmCapability, registerSemanticCapability, registerSemanticCapabilityWorker, registerServiceWorker, restoreDbSnapshot, searchTool, selectBackend, selectLlmModel, setEmbedFunction, setLlmFunction, sha256Hex, skosConceptToShard, skosRelationToShard, skosSchemeToShard, suggestTags, tagsFromShard, tagsToShard, titleGenerationHandler, unpackTarGz, unregisterLlmCapability, unregisterSemanticCapability, validateAiwgFortemiChunkManifest, validateAiwgFortemiChunkPart, validateAiwgFortemiIndexExport, validateAiwgStaticEmbeddingSet, validateChecksums, verifyDbSnapshotMeta, verifySri };
+export { AIWG_SCAN_REQUIRED_FIELDS, ArchiveManager, AttachmentsRepository, CURRENT_MIGRATION_HEAD, CURRENT_SHARD_VERSION, CapabilityManager, CaptureKnowledgeInputSchema, CollectionsRepository, CommunitiesRepository, DB_SNAPSHOT_SCHEMA_VERSION, DbSnapshotVersionError, EMBED_REQUEST_KIND, EMBED_RESPONSE_KIND, EmbeddingSetsRepository, FallbackRouter, FortemiToolManifest, GetNoteInputSchema, GraphRepository, JOB_CAPABILITIES, JOB_PRIORITIES, JobQueueWorker, LOCAL_ENDPOINTS, LinksRepository, ListNotesInputSchema, ManageArchiveInputSchema, ManageAttachmentsInputSchema, ManageCapabilitiesInputSchema, ManageCollectionsInputSchema, ManageLinksInputSchema, ManageNoteInputSchema, ManageTagsInputSchema, MemoryBlobStore, MigrationRunner, NotesRepository, OpenAICompatibleProvider, PGliteStorageBackend, PGliteStorageBackendFactory, PGliteWorkerClient, PGliteWorkerStorageBackend, PGliteWorkerStorageBackendFactory, ProvenanceRepository, ProviderRegistry, SHARD_FORMAT, SUPPORTED_PGLITE_VERSION, SearchInputSchema, SearchRepository, SkosRepository, TagsRepository, TransactionProxy, TypedEventBus, VERSION, aiRevisionHandler, aiwgFortemiIndexToCommunityGraph, allMigrations, appendPluginScript, assertAiwgFortemiChunkManifest, assertAiwgFortemiChunkPart, assertAiwgFortemiIndexExport, assertAiwgStaticEmbeddingSet, buildAiwgChunkedIndex, buildAiwgStaticEmbeddingSet, buildNoteConditions, buildPluginCsp, captureKnowledge, chunkText, classifyError, classifyModel, clearPrefetchedShard, collectionFromShard, collectionToShard, computeHash, computeSri, conceptTaggingHandler, cosineSimilarity, createAiwgFetchChunkLoader, createAiwgFetchDetailLoader, createAiwgIndexController, createAiwgReviewDecisionExport, createBlobStore, createCosineSemanticProvider, createCspReportHandler, createFortemi, createLegacyProvider, createPGliteBackend, createPGliteInstance, createRemoteBackend, createRoutes, createShardBackend, createWorkerEmbedFunction, defaultStorageBackendFactory, detectCommunities, detectGpuCapabilities, detectInferenceCapabilities, discoverLocalProviders, dumpDbSnapshot, embeddingFromShard, embeddingGenerationHandler, embeddingSetFromShard, embeddingSetMemberToShard, embeddingSetToShard, embeddingToShard, enqueueFullWorkflow, enqueueJob, enqueueNoteCreationJobs, estimateModelFit, estimateVramMB, estimateVramTier, exportShard, fetchPluginScript, findAiwgStaticDuplicatePairs, fortemiManifest, fromPrefetched, generateId, getAiwgFortemiFacets, getEmbedFunction, getFortemiBridge, getFortemiSecretStore, getJobQueueStatus, getLlmFunction, getNote, getPrefetchedSha256, handleEmbedRequests, hasFortemiSecureSecrets, importShard, isPluginScriptAllowed, isShardPrefetched, linkFromShard, linkToShard, linkingHandler, listNotes, manageArchive, manageAttachments, manageCapabilities, manageCollections, manageLinks, manageNote, manageTags, matchRoute, noteFromShard, noteSkosTagToShard, noteToShard, openShard, packTarGz, parseCspReport, prefetchShard, provenanceEdgeToShard, queryAiwgFortemiIndex, queryAiwgHybridIndex, queryAiwgSemanticIndex, registerLlmCapability, registerSemanticCapability, registerSemanticCapabilityWorker, registerServiceWorker, restoreDbSnapshot, searchTool, selectBackend, selectLlmModel, setEmbedFunction, setLlmFunction, sha256Hex, skosConceptToShard, skosRelationToShard, skosSchemeToShard, suggestTags, tagsFromShard, tagsToShard, titleGenerationHandler, unpackTarGz, unregisterLlmCapability, unregisterSemanticCapability, validateAiwgFortemiChunkManifest, validateAiwgFortemiChunkPart, validateAiwgFortemiIndexExport, validateAiwgStaticEmbeddingSet, validateChecksums, verifyDbSnapshotMeta, verifySri };
 //# sourceMappingURL=index.js.map
 //# sourceMappingURL=index.js.map