feat: workflow, UI and document support updates

backend/src/lib/chat/contextBuilders.ts

@@ -539,7 +539,7 @@ export async function buildWorkflowStore(
 
   // Seed system workflows first.
   for (const wf of SYSTEM_ASSISTANT_WORKFLOWS) {
-    store.set(wf.id, { title: wf.title, prompt_md: wf.prompt_md });
+    store.set(wf.id, { title: wf.title, skill_md: wf.skill_md });
   }
 
   // Then overlay user-owned assistant workflows.
@@ -550,7 +550,7 @@ export async function buildWorkflowStore(
     .eq("type", "assistant");
   for (const wf of workflows ?? []) {
     if (wf.prompt_md) {
-      store.set(wf.id, { title: wf.title, prompt_md: wf.prompt_md });
+      store.set(wf.id, { title: wf.title, skill_md: wf.prompt_md });
     }
   }
 
@@ -573,7 +573,7 @@ export async function buildWorkflowStore(
         if (wf.prompt_md) {
           store.set(wf.id, {
             title: wf.title,
-            prompt_md: wf.prompt_md,
+            skill_md: wf.prompt_md,
           });
         }
       }

backend/src/lib/chat/prompts.ts

@@ -32,6 +32,7 @@ Citation rules:
 - Keep quotes short, ideally 25 words or fewer, and tightly matched to the claim.
 - "page" means the sequential [Page N] marker in the provided text, not printed page numbers inside the document. Non-spreadsheet unpaginated files may have no [Page N] markers; omit "page" (or use 1) when none is present.
 - For spreadsheet sources (content shown as "## Sheet: <name>" markdown tables with a "Row" column and column-letter headers), cite by cell instead of page: set "sheet" to the sheet name and "cell" to the A1 address or range you are quoting (e.g. "B7" or "B7:C9", combining the column-letter header with the "Row" number). Put the plain cell value in "quote" with no "Row"/column-letter labels or "|" separators. Omit "page" for spreadsheet citations.
+- A cell tagged "⟨merged A1:C1⟩" spans that whole range: its value belongs to the anchor cell and the other covered cells are shown blank. When citing anything in a merged range, set "cell" to the full range from the tag (e.g. "A1:C1"), not a covered cell like "B1". Do not include the "⟨merged ...⟩" tag text in "quote".
 - For a continuous quote crossing two pages, set "page" to "N-M" and include [[PAGE_BREAK]] at the page break. Otherwise, use separate quote objects.
 - For legacy compatibility, you may also include top-level "page" and "quote" matching the first quote.
 - Omit the <CITATIONS> block when there are no citations.

backend/src/lib/chat/tools/toolDispatcher.ts

@@ -777,7 +777,7 @@ export async function runToolCalls(
       toolResults.push({
         role: "tool",
         tool_call_id: tc.id,
-        content: wf ? wf.prompt_md : `Workflow '${wfId}' not found.`,
+        content: wf ? wf.skill_md : `Workflow '${wfId}' not found.`,
       });
     } else if (tc.function.name === "read_table_cells" && tabularStore) {
       const colIndices = args.col_indices as number[] | undefined;

backend/src/lib/chat/types.ts

@@ -23,7 +23,7 @@ export type DocStore = Map<
   { storage_path: string; file_type: string; filename: string }
 >;
 
-export type WorkflowStore = Map<string, { title: string; prompt_md: string }>;
+export type WorkflowStore = Map<string, { title: string; skill_md: string }>;
 
 export type DocIndex = Record<
   string,

backend/src/lib/spreadsheet.ts

@@ -33,6 +33,16 @@ function renderSheet(sheetName: string, ws: XLSX.WorkSheet): string | null {
   if (!ref) return null;
   const range = XLSX.utils.decode_range(ref);
 
+  // Map each merged range's top-left (anchor) address to its encoded range so we
+  // can tag the anchor inline (e.g. `Amount ⟨merged B2:C2⟩`). The covered cells
+  // stay blank, so the model never reads a covered address (e.g. B1 inside
+  // A1:C1) as its own value; the tag tells it the anchor spans that range, and
+  // to cite the whole range for anything in it.
+  const mergeAnchors = new Map<string, string>();
+  for (const m of ws["!merges"] ?? []) {
+    mergeAnchors.set(XLSX.utils.encode_cell(m.s), XLSX.utils.encode_range(m));
+  }
+
   // Build a trimmed grid: capture formatted text for every cell in the used
   // range, then drop trailing empty columns and fully empty rows so we don't
   // emit oceans of blank cells.
@@ -44,7 +54,13 @@ function renderSheet(sheetName: string, ws: XLSX.WorkSheet): string | null {
     let rowHasContent = false;
     for (let c = range.s.c; c <= range.e.c; c++) {
       const addr = XLSX.utils.encode_cell({ r, c });
-      const text = sanitizeCellText(cellDisplayText(ws[addr]));
+      let text = sanitizeCellText(cellDisplayText(ws[addr]));
+      const mergeRange = mergeAnchors.get(addr);
+      if (mergeRange) {
+        text = text
+          ? `${text} ⟨merged ${mergeRange}⟩`
+          : `⟨merged ${mergeRange}⟩`;
+      }
       cells[c - range.s.c] = text;
       if (text) {
         rowHasContent = true;
@@ -72,13 +88,6 @@ function renderSheet(sheetName: string, ws: XLSX.WorkSheet): string | null {
 
   const lines = [`## Sheet: ${sheetName}`, "", headerRow, separator, ...bodyRows];
 
-  // Note merged ranges once so the model understands spanned headers/labels.
-  const merges = ws["!merges"];
-  if (merges && merges.length > 0) {
-    const encoded = merges.map((m) => XLSX.utils.encode_range(m));
-    lines.push("", `Merged ranges: ${encoded.join(", ")}`);
-  }
-
   return lines.join("\n");
 }
 

backend/src/lib/userLookup.ts

@@ -0,0 +1,113 @@
+import type { SupabaseClient } from "@supabase/supabase-js";
+
+type Db = SupabaseClient<any, "public", any>;
+
+export type ProfileUserInfo = {
+    id: string;
+    email: string;
+    display_name: string | null;
+};
+
+export function normalizeEmail(value: unknown) {
+    return typeof value === "string" ? value.trim().toLowerCase() : "";
+}
+
+export function normalizeDisplayName(value: unknown) {
+    return typeof value === "string" && value.trim() ? value.trim() : null;
+}
+
+export async function loadProfileUsersByEmail(db: Db) {
+    const { data, error } = await db
+        .from("user_profiles")
+        .select("user_id, email, display_name")
+        .not("email", "is", null);
+    if (error) throw error;
+
+    const userByEmail = new Map<string, ProfileUserInfo>();
+    const userById = new Map<string, ProfileUserInfo>();
+    for (const row of data ?? []) {
+        const email = normalizeEmail(row.email);
+        if (!email) continue;
+        const info = {
+            id: row.user_id as string,
+            email,
+            display_name: normalizeDisplayName(row.display_name),
+        };
+        userByEmail.set(email, info);
+        userById.set(info.id, info);
+    }
+
+    return { userByEmail, userById };
+}
+
+export async function findProfileUserByEmail(db: Db, email: string) {
+    const normalized = normalizeEmail(email);
+    if (!normalized) return null;
+
+    const { data, error } = await db
+        .from("user_profiles")
+        .select("user_id, email, display_name")
+        .eq("email", normalized)
+        .maybeSingle();
+    if (error) throw error;
+    if (!data) return null;
+
+    return {
+        id: data.user_id as string,
+        email: normalized,
+        display_name: normalizeDisplayName(data.display_name),
+    };
+}
+
+export async function findMissingUserEmails(db: Db, emails: string[]) {
+    const normalizedEmails = [...new Set(emails.map(normalizeEmail).filter(Boolean))];
+    if (normalizedEmails.length === 0) return [];
+
+    const { data, error } = await db
+        .from("user_profiles")
+        .select("email")
+        .in("email", normalizedEmails);
+    if (error) throw error;
+
+    const found = new Set(
+        (data ?? [])
+            .map((row) => normalizeEmail(row.email))
+            .filter(Boolean),
+    );
+    return normalizedEmails.filter((email) => !found.has(email));
+}
+
+export async function syncProfileEmail(
+    db: Db,
+    userId: string,
+    email: string | null | undefined,
+) {
+    const normalizedEmail = normalizeEmail(email);
+    if (!userId || !normalizedEmail) return null;
+
+    const { data: existing, error: loadError } = await db
+        .from("user_profiles")
+        .select("email")
+        .eq("user_id", userId)
+        .maybeSingle();
+    if (loadError) return loadError;
+
+    if (!existing) {
+        const { error } = await db.from("user_profiles").insert({
+            user_id: userId,
+            email: normalizedEmail,
+        });
+        return error;
+    }
+
+    if (normalizeEmail(existing.email) === normalizedEmail) return null;
+
+    const { error } = await db
+        .from("user_profiles")
+        .update({
+            email: normalizedEmail,
+            updated_at: new Date().toISOString(),
+        })
+        .eq("user_id", userId);
+    return error;
+}

backend/src/routes/workflows.ts

@@ -32,12 +32,25 @@ type WorkflowRecord = {
   [key: string]: unknown;
 };
 
+type WorkflowType = "assistant" | "tabular";
+
 type WorkflowContributor = {
   name: string;
   organisation: string | null;
   role: string | null;
   linkedin: string | null;
 };
+
+type WorkflowMetadata = {
+  title: string;
+  description: string | null;
+  type: WorkflowType;
+  contributors: WorkflowContributor[];
+  language: string;
+  version: string | null;
+  practice: string | null;
+  jurisdictions: string[] | null;
+};
 type OpenSourceSubmissionStatus = "pending" | "approved" | "rejected";
 
 type OpenSourceSubmissionRow = {
@@ -52,7 +65,6 @@ type OpenSourceSubmissionRow = {
   submitted_at: string;
   updated_at: string;
   reviewed_at?: string | null;
-  reviewed_by_user_id?: string | null;
   review_notes?: string | null;
 };
 
@@ -120,9 +132,42 @@ function withSystemWorkflowAccess(workflow: SystemWorkflow) {
   });
 }
 
-function withDatabaseWorkflow<T extends object>(workflow: T) {
+function workflowTypeFrom(value: unknown): WorkflowType {
+  return value === "tabular" ? "tabular" : "assistant";
+}
+
+function metadataFromWorkflowRecord(workflow: WorkflowRecord): WorkflowMetadata {
   return {
-    ...workflow,
+    title: workflow.title ?? "",
+    description: null,
+    type: workflowTypeFrom(workflow.type),
+    contributors:
+      normalizeContributors(workflow.contributors) ?? [
+        DEFAULT_WORKFLOW_CONTRIBUTOR,
+      ],
+    language: workflow.language ?? DEFAULT_WORKFLOW_LANGUAGE,
+    version: workflow.version ?? null,
+    practice: workflow.practice ?? DEFAULT_WORKFLOW_PRACTICE,
+    jurisdictions: workflow.jurisdictions ?? DEFAULT_WORKFLOW_JURISDICTIONS,
+  };
+}
+
+function withDatabaseWorkflow(workflow: WorkflowRecord) {
+  const {
+    title: _title,
+    type: _type,
+    contributors: _contributors,
+    language: _language,
+    version: _version,
+    practice: _practice,
+    jurisdictions: _jurisdictions,
+    prompt_md,
+    ...rest
+  } = workflow;
+  return {
+    ...rest,
+    metadata: metadataFromWorkflowRecord(workflow),
+    skill_md: prompt_md ?? null,
     is_system: false,
   };
 }
@@ -179,7 +224,7 @@ async function resolveWorkflowAccess(
     .eq("id", workflowId)
     .single();
   if (!workflow) return null;
-  const workflowRecord = withDatabaseWorkflow(workflow as WorkflowRecord);
+  const workflowRecord = workflow as WorkflowRecord;
   if (workflowRecord.user_id === userId) {
     return { workflow: workflowRecord, allowEdit: true, isOwner: true };
   }
@@ -234,16 +279,13 @@ function buildOpenSourceSnapshot(
 ) {
   return {
     workflow_id: workflow.id,
-    title: workflow.title ?? "",
-    type: workflow.type ?? "",
-    prompt_md: workflow.prompt_md ?? null,
+    metadata: {
+      ...metadataFromWorkflowRecord(workflow),
+      contributors,
+    },
+    skill_md: workflow.prompt_md ?? null,
     columns_config: workflow.columns_config ?? null,
-    contributors,
     contributor_mode: contributorMode,
-    language: workflow.language ?? DEFAULT_WORKFLOW_LANGUAGE,
-    version: workflow.version ?? null,
-    practice: workflow.practice ?? DEFAULT_WORKFLOW_PRACTICE,
-    jurisdictions: workflow.jurisdictions ?? DEFAULT_WORKFLOW_JURISDICTIONS,
     created_at: workflow.created_at ?? null,
   };
 }
@@ -280,11 +322,11 @@ workflowsRouter.get("/", requireAuth, asyncRoute(async (req, res) => {
   }
 
   const systemWorkflows = SYSTEM_WORKFLOWS.filter(
-    (workflow) => !workflowType || workflow.type === workflowType,
+    (workflow) => !workflowType || workflow.metadata.type === workflowType,
   ).map(withSystemWorkflowAccess);
   const databaseWorkflows = ((data ?? []) as WorkflowRecord[]).filter(
     (workflow) => !SYSTEM_WORKFLOW_IDS.has(workflow.id),
-  );
+  ).map(withDatabaseWorkflow);
 
   res.json([...systemWorkflows, ...databaseWorkflows]);
 }));
@@ -293,40 +335,36 @@ workflowsRouter.get("/", requireAuth, asyncRoute(async (req, res) => {
 workflowsRouter.post("/", requireAuth, asyncRoute(async (req, res) => {
   const userId = res.locals.userId as string;
   const {
-    title,
-    type,
-    prompt_md,
+    metadata,
+    skill_md,
     columns_config,
-    language,
-    practice,
-    jurisdictions,
   } = req.body as {
-    title: string;
-    type: string;
-    prompt_md?: string;
+    metadata?: Partial<WorkflowMetadata>;
+    skill_md?: string;
     columns_config?: unknown;
-    language?: unknown;
-    practice?: string | null;
-    jurisdictions?: unknown;
   };
+  const title = metadata?.title;
+  const type = metadata?.type;
   if (!title?.trim())
-    return void res.status(400).json({ detail: "title is required" });
-  if (!["assistant", "tabular"].includes(type))
+    return void res.status(400).json({ detail: "metadata.title is required" });
+  if (type !== "assistant" && type !== "tabular")
     return void res
       .status(400)
-      .json({ detail: "type must be 'assistant' or 'tabular'" });
+      .json({ detail: "metadata.type must be 'assistant' or 'tabular'" });
 
   const db = createServerSupabase();
   devLog("[workflows/create] request", {
     userId,
     title: title.trim(),
     type,
-    hasPrompt: typeof prompt_md === "string" && prompt_md.length > 0,
+    hasSkill: typeof skill_md === "string" && skill_md.length > 0,
     columnCount: Array.isArray(columns_config) ? columns_config.length : null,
-    language: normalizeOptionalString(language) ?? DEFAULT_WORKFLOW_LANGUAGE,
-    practice: practice ?? null,
+    language:
+      normalizeOptionalString(metadata?.language) ?? DEFAULT_WORKFLOW_LANGUAGE,
+    practice: metadata?.practice ?? null,
     jurisdictions:
-      normalizeJurisdictions(jurisdictions) ?? DEFAULT_WORKFLOW_JURISDICTIONS,
+      normalizeJurisdictions(metadata?.jurisdictions) ??
+      DEFAULT_WORKFLOW_JURISDICTIONS,
   });
   const { data, error } = await db
     .from("workflows")
@@ -334,13 +372,15 @@ workflowsRouter.post("/", requireAuth, asyncRoute(async (req, res) => {
       user_id: userId,
       title: title.trim(),
       type,
-      prompt_md: prompt_md ?? null,
+      prompt_md: skill_md ?? null,
       columns_config: columns_config ?? null,
-      language: normalizeOptionalString(language) ?? DEFAULT_WORKFLOW_LANGUAGE,
+      language:
+        normalizeOptionalString(metadata?.language) ?? DEFAULT_WORKFLOW_LANGUAGE,
       practice:
-        normalizeOptionalString(practice) ?? DEFAULT_WORKFLOW_PRACTICE,
+        normalizeOptionalString(metadata?.practice) ?? DEFAULT_WORKFLOW_PRACTICE,
       jurisdictions:
-        normalizeJurisdictions(jurisdictions) ?? DEFAULT_WORKFLOW_JURISDICTIONS,
+        normalizeJurisdictions(metadata?.jurisdictions) ??
+        DEFAULT_WORKFLOW_JURISDICTIONS,
     })
     .select("*")
     .single();
@@ -362,7 +402,7 @@ workflowsRouter.post("/", requireAuth, asyncRoute(async (req, res) => {
     title: data?.title,
     type: data?.type,
   });
-  res.status(201).json(withDatabaseWorkflow(data));
+  res.status(201).json(withDatabaseWorkflow(data as WorkflowRecord));
 }));
 
 async function handleWorkflowUpdate(req: Request, res: Response) {
@@ -370,15 +410,17 @@ async function handleWorkflowUpdate(req: Request, res: Response) {
   const userEmail = res.locals.userEmail as string | undefined;
   const { workflowId } = req.params;
   const updates: Record<string, unknown> = {};
-  if (req.body.title != null) updates.title = req.body.title;
-  if (req.body.prompt_md != null) updates.prompt_md = req.body.prompt_md;
+  const metadata = req.body.metadata as Partial<WorkflowMetadata> | undefined;
+  if (metadata?.title != null) updates.title = metadata.title;
+  if (req.body.skill_md != null) updates.prompt_md = req.body.skill_md;
   if (req.body.columns_config != null)
     updates.columns_config = req.body.columns_config;
-  if ("language" in req.body)
-    updates.language = normalizeOptionalString(req.body.language);
-  if ("practice" in req.body) updates.practice = req.body.practice ?? null;
-  if ("jurisdictions" in req.body)
-    updates.jurisdictions = normalizeJurisdictions(req.body.jurisdictions);
+  if (metadata && "language" in metadata)
+    updates.language = normalizeOptionalString(metadata.language);
+  if (metadata && "practice" in metadata)
+    updates.practice = metadata.practice ?? null;
+  if (metadata && "jurisdictions" in metadata)
+    updates.jurisdictions = normalizeJurisdictions(metadata.jurisdictions);
 
   const db = createServerSupabase();
   const access = await resolveWorkflowAccess(workflowId, userId, userEmail, db);
@@ -398,7 +440,7 @@ async function handleWorkflowUpdate(req: Request, res: Response) {
       .status(404)
       .json({ detail: "Workflow not found or not editable" });
   res.json(
-    withWorkflowAccess(withDatabaseWorkflow(data), {
+    withWorkflowAccess(withDatabaseWorkflow(data as WorkflowRecord), {
       allowEdit: access.allowEdit,
       isOwner: access.isOwner,
     }),
@@ -506,7 +548,7 @@ workflowsRouter.post("/:workflowId/open-source", requireAuth, asyncRoute(async (
       .json({ detail: "Workflow not found or not open-sourceable" });
   }
 
-  const workflowRecord = withDatabaseWorkflow(workflow as WorkflowRecord);
+  const workflowRecord = workflow as WorkflowRecord;
   const validationError = validateOpenSourceWorkflow(workflowRecord);
   if (validationError) {
     return void res.status(400).json({ detail: validationError });
@@ -620,7 +662,7 @@ workflowsRouter.get("/:workflowId", requireAuth, asyncRoute(async (req, res) =>
     : null;
   res.json(
     withOpenSourceSubmission(
-      withWorkflowAccess(access.workflow, {
+      withWorkflowAccess(withDatabaseWorkflow(access.workflow), {
         allowEdit: access.allowEdit,
         isOwner: access.isOwner,
       }),