aditya-2k23 · aditya-2k23 · Apr 30, 2026 · Apr 30, 2026 · Apr 30, 2026 · Apr 30, 2026
diff --git a/.gitignore b/.gitignore
@@ -43,3 +43,5 @@ yarn-error.log*
 next-env.d.ts
 
 .vscode
+# Ignore Git Commit Genie data
+.gitgenie/**
diff --git a/AGENT.md b/AGENT.md
@@ -37,12 +37,6 @@ Users log:
 - **AI**: Google Gemini (@google/generative-ai SDK)
 - **Language**: JavaScript
 
-Do NOT:
-
-- Introduce Redux, Zustand, or new state libraries
-- Introduce new backend frameworks
-- Add experimental dependencies without strong reason
-
 ---
 
 ## 📁 Project Structure (Mental Model)
@@ -133,7 +127,6 @@ When fixing bugs:
 
 Agents must NOT:
 
-- Rewrite large files without request
 - Change formatting or lint rules arbitrarily
 - Rename variables for “style”
 - Add features unless explicitly asked

diff --git a/app/actions/insights.js b/app/actions/insights.js
@@ -242,6 +242,9 @@ function buildPrompt(journalEntry) {
   - You NEVER sound like a therapist, a bot, or a report. You sound like a girl who genuinely cares and is reading their journal.
   - These phrases are banned forever: "I hear you", "that's valid", "it sounds like", "it's okay to feel", "as an AI", "I understand that", "I notice a pattern"
 
+  ABOUT THE JOURNAL:
+  The user writes their journal entries using a rich text editor that supports formatting — bold, italics, headings (H1, H2), and blockquotes. Pay attention to how they've formatted their entry. Bold text usually signals something they want to emphasize. Italics often carry softer, more reflective thoughts. Headings may structure a longer entry into sections. Blockquotes may be something they've heard, remembered, or want to reflect on. Let these formatting cues shape your understanding of what matters most to them today.
+
   JOURNAL ENTRY:
   """
   ${journalEntry}
@@ -255,18 +258,21 @@ function buildPrompt(journalEntry) {
   2. TRIGGERS — 2 to 4 short phrases (1-3 words each):
   - Pulled directly from what they actually wrote
   - Specific events, people, or themes mentioned — not generic labels
+  - If something is bolded or emphasized, it's likely a trigger worth calling out
   - Examples: "missed deadline", "toxic coworker", "no sleep", "good news from mom"
 
   3. RESPONSE — a warm, personal paragraph (3-5 sentences) written like a best friend reacting to their journal:
   - Open with your immediate emotional reaction to what they shared — make it feel real, not scripted
-  - Mention something specific from their entry so they know you actually read it
+  - Mention something specific from their entry so they know you actually read it — especially anything they emphasized with bold or italic
+  - Make sure you also, use the ability of the text editor formatting to show that you see the nuances in their feelings.
   - Celebrate the win OR sit with them in the hard feeling — don't rush past either
   - Keep it warm, a little conversational, sprinkle emojis naturally where a person actually would 🥺
   - NO bullet points, NO lists — just natural flowing prose like a friend texting a longer message
   - Do NOT end with a question here — the follow-up question is separate
 
   4. FOCUS — one specific thing from their entry (a problem OR a positive moment) that you're zooming in on:
   - Should be a short phrase, not a sentence
+  - Prioritize something they formatted with emphasis if it exists — that formatting is intentional
   - This is what the follow-up question will be anchored to
 
   5. FOLLOW-UP QUESTION — one question that flows naturally from the focus:
@@ -292,6 +298,9 @@ function buildPartialPrompt(journalEntry, cachedMood, cachedTriggers, cachedHead
   - Things on their mind lately: ${cachedTriggers.join(", ")}
   - Last headline you gave them: "${cachedHeadline}"
 
+  ABOUT THE JOURNAL:
+  The user writes their journal entries using a rich text editor that supports bold, italics, headings (H1, H2), and blockquotes. If their entry includes formatted text, treat it as intentional emphasis. Bold usually signals something important or emotionally charged. Italics often carry softer reflections or hesitant thoughts. Headings may segment a longer entry into distinct moments. Blockquotes may be something they heard, remembered, or want to sit with. Let these cues guide where you focus your response.
+
   WHAT THEY WROTE TODAY:
   """
   ${journalEntry}
@@ -301,14 +310,16 @@ function buildPartialPrompt(journalEntry, cachedMood, cachedTriggers, cachedHead
   1. RESPONSE — a warm, personal paragraph (3-5 sentences) written like a best friend reacting to today's entry:
   - Gently acknowledge that this feeling or situation has been coming up — but do it warmly, like a friend who notices and cares, not like a system detecting a pattern
   - Example tone: "hey, this keeps coming up and I just wanna make sure you're okay 🥺" — NOT "I notice a recurring pattern in your entries"
-  - React to something specific in TODAY's entry — show you read this one, not just the last
+  - React to something specific in TODAY's entry — show you read this one, not just the last. If they bolded or italicized something, that's worth acknowledging
+  - Use the formatting cues to show you see the nuances in their feelings today.
   - Keep it warm, conversational, with emojis where they naturally fit
   - NO bullet points, NO lists — flowing prose only
   - Do NOT end with a question here
 
   2. FOCUS — one specific thing from today's entry to zoom in on:
   - Short phrase, not a sentence
   - Should be something slightly different from the previous focus to help them explore a new angle
+  - Formatted or emphasized text from their entry is often the best place to look
 
   3. FOLLOW-UP QUESTION — one fresh question from a new angle:
   - Try a different angle from what you asked before — help them explore something they haven't said yet

diff --git a/app/api/chat/history/route.js b/app/api/chat/history/route.js
@@ -65,9 +65,13 @@ export async function GET(req) {
       const sId = data.sessionId || "default";
 
       if (!groupedSessions[sId]) {
+        let cleanContent = data.content || "";
+        if (typeof cleanContent === "string") {
+            cleanContent = cleanContent.trim().replace(/^["']+|["']+$/g, "");
+        }
         groupedSessions[sId] = {
           sessionId: sId,
-          preview: (data.content || "").substring(0, 40) + ((data.content || "").length > 40 ? '...' : ''),
+          preview: cleanContent.substring(0, 40) + (cleanContent.length > 40 ? '...' : ''),
           messages: []
         };
       }
@@ -79,15 +83,19 @@ export async function GET(req) {
             groupedSessions[sId].messages.push({
               id: `${doc.id}_${index}`,
               role: "assistant",
-              content: bubble,
+              content: bubble.trim().replace(/^["']|["']$/g, ""),
               timestamp,
             });
           });
       } else {
+        let cleanContent = data.content;
+        if (typeof cleanContent === "string") {
+            cleanContent = cleanContent.trim().replace(/^["']+|["']+$/g, "");
+        }
         groupedSessions[sId].messages.push({
           id: doc.id,
           role: data.role,
-          content: data.content,
+          content: cleanContent,
           timestamp,
         });
       }