fix: catch the more common tool-skip case — short reply, no tool call (v0.57.2)

The previous fix's looksLikeUnstructuredRecipe only fires when the model
spells the whole recipe out as a numbered/bulleted list — it does nothing
for a short reply that just *claims* a draft exists ("Voici une recette
de Bœuf Bourguignon... brouillon à vérifier ci-dessous") without ever
calling the tool, which is what the user actually hit.

Detecting the model's claimed compliance is fragile (varies by phrasing/
model), so instead detect intent from the user's own message — the same
noun+verb pattern ("recipe"+create/make/... or "recette"+créer/fais/...)
the system prompt's own createRecipe examples already describe. Forces
the retry whenever there's no tool call AND either signal fires.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

CHANGELOG.md

@@ -2,6 +2,11 @@
 
 All notable changes to Epicure are documented here. This file is mirrored in-app at `/changelog` (and in the admin dashboard) via `apps/web/lib/changelog.ts` — update both together.
 
+## 0.57.2 — 2026-07-20 20:15
+
+### Fixed
+- The previous chatbot fix only caught the model spelling a recipe out as prose. The more common case: a short reply that claims a draft exists ("here's a draft, check below") without ever calling the tool, so nothing renders. Now also detects recipe-creation intent from the user's own message and forces the tool call in that case too.
+
 ## 0.57.1 — 2026-07-20 20:00
 
 ### Fixed

apps/web/app/api/v1/ai/cooking-chat/route.ts

@@ -37,6 +37,24 @@ function looksLikeUnstructuredRecipe(text: string): boolean {
   return listLines.length >= 3;
 }
 
+// The above only catches the model spelling the whole recipe out. A second,
+// more common failure with weaker models: it gives a short reply that even
+// *claims* to have drafted something ("here's a draft, check below") without
+// ever calling the tool — so nothing renders below it. Rather than guess from
+// that claim (fragile across models/phrasing), detect intent from the user's
+// own message instead: same noun+verb pattern the system prompt itself uses
+// as its createRecipe examples, checked in whichever of the two supported
+// locales the request is in.
+const RECIPE_INTENT: Record<string, { noun: RegExp; verb: RegExp }> = {
+  en: { noun: /\brecipe\b/i, verb: /\b(create|make|generate|give|write|save|invent)\b/i },
+  fr: { noun: /\brecettes?\b/i, verb: /\b(cr[ée]e?r?|fais|donne|[ée]cri[st]|note[rz]?|sauvegard\w*|enregistr\w*|invente\w*|g[ée]n[èe]re\w*)\b/i },
+};
+
+function looksLikeRecipeCreationRequest(question: string, locale: string): boolean {
+  const pattern = RECIPE_INTENT[locale] ?? RECIPE_INTENT["en"]!;
+  return pattern.noun.test(question) && pattern.verb.test(question);
+}
+
 export async function POST(req: NextRequest) {
   const { session, response } = await requireSessionOrApiKey(req);
   if (response) return response;
@@ -77,10 +95,14 @@ export async function POST(req: NextRequest) {
   let final = result.data;
 
   // Weaker/free-tier models sometimes ignore the MUST-call-the-tool rule
-  // above and just write the recipe out in prose instead. Rather than
-  // surface that (or fail the request), force the tool call on one extra
-  // pass — same system+prompt, just no longer letting the model opt out.
-  if (final.toolCalls.length === 0 && looksLikeUnstructuredRecipe(final.text)) {
+  // above — either by spelling the recipe out in prose, or (more common,
+  // and easy to miss) giving a short reply that *claims* a draft exists
+  // without ever calling the tool, so nothing actually renders below it.
+  // Catch both: the prose-list shape, or the user's own message clearly
+  // asking for a recipe in the first place. Rather than surface either
+  // failure, force the tool call on one extra pass — same system+prompt,
+  // just no longer letting the model opt out.
+  if (final.toolCalls.length === 0 && (looksLikeUnstructuredRecipe(final.text) || looksLikeRecipeCreationRequest(parsed.data.question, locale))) {
     try {
       const forced = await generateText({
         model,

apps/web/lib/changelog.ts

@@ -1,5 +1,5 @@
 // Mirrors CHANGELOG.md at the repo root — update both together.
-export const APP_VERSION = "0.57.1";
+export const APP_VERSION = "0.57.2";
 
 export type ChangelogEntry = {
   version: string;
@@ -11,6 +11,13 @@ export type ChangelogEntry = {
 };
 
 export const CHANGELOG: ChangelogEntry[] = [
+  {
+    version: "0.57.2",
+    date: "2026-07-20 20:15",
+    fixed: [
+      "The previous chatbot fix only caught the model spelling a recipe out as prose. The more common case: a short reply that claims a draft exists (\"here's a draft, check below\") without ever calling the tool, so nothing renders. Now also detects recipe-creation intent from the user's own message and forces the tool call in that case too.",
+    ],
+  },
   {
     version: "0.57.1",
     date: "2026-07-20 20:00",

apps/web/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@epicure/web",
-  "version": "0.57.1",
+  "version": "0.57.2",
   "private": true,
   "scripts": {
     "dev": "next dev",

package.json

@@ -1,6 +1,6 @@
 {
   "name": "epicure",
-  "version": "0.57.1",
+  "version": "0.57.2",
   "private": true,
   "scripts": {
     "dev": "pnpm --filter web dev",