diff --git a/memento-note/lib/ai/services/agent-executor.service.ts b/memento-note/lib/ai/services/agent-executor.service.ts
index 4e09d2c..c189ac3 100644
--- a/memento-note/lib/ai/services/agent-executor.service.ts
+++ b/memento-note/lib/ai/services/agent-executor.service.ts
@@ -945,7 +945,7 @@ Appelle generate_slides avec un objet JSON structuré :
 12. "summary"    → { type:"summary", title:"...", items:["Point clé 1", "Point clé 2", ...] }
 
 ═══ RÈGLES ═══
-- 6 à 12 slides par présentation
+- Nombre de slides : adapté au contenu réel de la note (la règle ⚠️ dans le message utilisateur est ABSOLUE)
 - Slide 1 OBLIGATOIREMENT type "title"
 - Dernière slide OBLIGATOIREMENT type "summary"
 - Au moins 1 slide "chart" ou "stats" si des chiffres existent dans la note
@@ -989,7 +989,7 @@ Call generate_slides with a structured JSON object:
 12. "summary"    → { type:"summary", title:"...", items:["Key point 1", "Key point 2", ...] }
 
 ═══ RULES ═══
-- 6 to 12 slides per presentation
+- Slide count: adapt to the real content (the ⚠️ rule in the user message is ABSOLUTE)
 - Slide 1 MUST be type "title"
 - Last slide MUST be type "summary"
 - At least 1 "chart" or "stats" slide if numbers exist in the note
@@ -1027,6 +1027,49 @@ You MUST use the task_extract tool. Do NOT respond with text, call the tool dire
   },
 }
 
+/**
+ * Strip markdown syntax, URLs, metadata blocks, and frontmatter from note content.
+ * Returns only the semantic plain text, used to count real words before generating slides.
+ */
+function stripNoteMarkdown(raw: string): string {
+  return raw
+    // Remove frontmatter / YAML blocks
+    .replace(/^---[\s\S]*?---/m, '')
+    // Remove markdown headers
+    .replace(/^#{1,6}\s+/gm, '')
+    // Remove bold/italic
+    .replace(/[*_]{1,3}([^*_]+)[*_]{1,3}/g, '$1')
+    // Remove inline links — keep label only
+    .replace(/\[([^\]]+)\]\([^)]+\)/g, '$1')
+    // Remove bare URLs
+    .replace(/https?:\/\/\S+/g, '')
+    // Remove metadata lines (Connection to seed, Novelty score, Source brainstorm, Origin, Derived from, etc.)
+    .replace(/^\*{0,2}(Connection to seed|Novelty score|Source brainstorm|Source note|Origin|Derived from|## Origin)[^:\n]*:?.*$/gim, '')
+    // Remove horizontal rules
+    .replace(/^---+$/gm, '')
+    // Remove blockquote markers
+    .replace(/^>\s*/gm, '')
+    // Remove code blocks
+    .replace(/```[\s\S]*?```/g, '')
+    .replace(/`[^`]+`/g, '')
+    // Collapse whitespace
+    .replace(/\s+/g, ' ')
+    .trim()
+}
+
+/** Count real semantic words in a note */
+function countNoteWords(raw: string): number {
+  return stripNoteMarkdown(raw).split(/\s+/).filter(Boolean).length
+}
+
+/** Return max slides based on real word count */
+function slideLimit(wordCount: number): { max: number; label: string } {
+  if (wordCount < 50)  return { max: 3, label: '3 slides MAX (note très courte)' }
+  if (wordCount < 150) return { max: 4, label: '4 slides MAX (note courte)' }
+  if (wordCount < 350) return { max: 6, label: '6 slides MAX (note moyenne)' }
+  return { max: 8, label: '8 slides MAX (note longue)' }
+}
+
 function extractJsonFromText(text: string): any {
   if (!text) return null
 
@@ -1274,7 +1317,16 @@ async function executeToolUseAgent(
           const notesContext = notes.map(n =>
             `### ${n.title || untitled} (${n.createdAt.toLocaleDateString(dateLocale)})\n${n.content.substring(0, 2000)}`
           ).join('\n\n')
+
+          // Compute real word count (stripped of markdown/metadata) to determine slide limit
+          const totalWords = notes.reduce((sum, n) => sum + countNoteWords(n.content || ''), 0)
+          const { max: maxSlides, label: slideRule } = slideLimit(totalWords)
+
           prompt += `\n\n${lang === 'fr' ? 'Notes source' : 'Source notes'}:\n\n${notesContext}`
+          prompt += `\n\n${lang === 'fr'
+            ? `⚠️ RÈGLE ABSOLUE : Cette note fait ${totalWords} mots réels (hors markdown/URLs/métadonnées). Tu DOIS générer EXACTEMENT ${slideRule}. Respecter cette limite est PLUS IMPORTANT que la diversité ou la densité.`
+            : `⚠️ ABSOLUTE RULE: This note has ${totalWords} real words (excluding markdown/URLs/metadata). You MUST generate EXACTLY ${slideRule}. This limit is MORE IMPORTANT than diversity or density.`
+          }`
         }
 
         // ── Executive Template Structure (HTML-compatible) ──
@@ -1317,8 +1369,8 @@ async function executeToolUseAgent(
           : `\n\nMANDATORY DENSITY: Every slide must have RICH CONTENT. Minimum 5 items per "bullets" slide (≥15 words each). Minimum 4 cards when using "cards". Minimum 4 data points for "chart". You MUST include at least 1 "chart" slide with REAL data from the notes. If no numbers in notes, create a maturity radar or qualitative comparison scored out of 5.`
 
         prompt += `\n\n${lang === 'fr'
-          ? 'IMPORTANT : Appelle OBLIGATOIREMENT generate_slides avec le JSON structuré {title, theme, slides:[...]}. Ne réponds JAMAIS avec du texte brut. 6-12 slides variées.'
-          : 'IMPORTANT: You MUST call generate_slides with structured JSON {title, theme, slides:[...]}. NEVER respond with plain text. 6-12 varied slides.'}`
+          ? 'IMPORTANT : Appelle OBLIGATOIREMENT generate_slides avec le JSON structuré {title, theme, slides:[...]}. Ne réponds JAMAIS avec du texte brut. Respecte ABSOLUMENT la limite de slides indiquée.'
+          : 'IMPORTANT: You MUST call generate_slides with structured JSON {title, theme, slides:[...]}. NEVER respond with plain text. STRICTLY respect the slide count limit indicated.'}`
 
         if (agent.slideTheme && agent.slideTheme !== 'auto') {
           prompt += `\n${lang === 'fr'