fix(chat): share calm voice character rules with notifications

src/Orbit.Infrastructure/Services/Prompts/NotificationVoice.cs

@@ -2,12 +2,17 @@ namespace Orbit.Infrastructure.Services.Prompts;
 
 public static class NotificationVoice
 {
-    public const string Rules = """
-        Voice, and every line here is absolute:
-        - Write calmly. Describe what is true. Never sell, never hype, never perform enthusiasm.
+    public const string CharacterRules = """
+        Character bans are absolute in every language:
         - Use no exclamation mark anywhere, in the title or in the body.
         - Use no dash character: no em dash, no en dash, and no doubled hyphen. A comma or a full stop does that job.
-        - Use no emoji, no markdown, no bullet, no heading, no greeting, no sign-off, and no quotation marks around the message.
+        - Use no emoji in the reply's own words.
+        """;
+
+    public const string Rules = CharacterRules + "\n" + """
+        Voice, and every line here is absolute:
+        - Write calmly. Describe what is true. Never sell, never hype, never perform enthusiasm.
+        - Use no markdown, no bullet, no heading, no greeting, no sign-off, and no quotation marks around the message.
         - Never shame, blame, scold or warn. A missed day is ordinary and is never a failure.
         - Address the reader directly as you. Never call them the user.
         - Use plain verbs. Never use these words or their translations: unlock, elevate, empower, supercharge, leverage, harness, delve, seamless, robust, crucial, essential, vital, revolutionary, maximize, optimize.

src/Orbit.Infrastructure/Services/Prompts/Sections/Dynamic/ImageInstructionsSection.cs

@@ -25,7 +25,7 @@ When the user uploads an image (photo of schedule, to-do list, calendar, task ap
                - Top-level groups/categories -> separate CreateHabit actions (parents)
                - Nested/indented items under a group -> subHabits array on that parent
                - Deeply nested items (sub-sub-items) -> flatten into the nearest parent's subHabits
-            3. Preserve exact titles/names as shown in the image -- do not rename, summarize, or merge items
+            3. Preserve exact titles/names as shown in the image. Do not rename, summarize, or merge items
             4. If an item appears multiple times (e.g., "Water - 710ml" x4), create each one individually
             5. Items with no children that are not nested under anything -> standalone CreateHabit (no subHabits)
             6. Infer frequency from visual cues:
@@ -45,7 +45,7 @@ 4. If an item appears multiple times (e.g., "Water - 710ml" x4), create each one
 
             ### Example: Task app screenshot with groups (user says "create these habits")
             {
-              "aiMessage": "Created 5 habit groups with all their sub-habits from your screenshot!",
+              "aiMessage": "Created 5 habit groups with all their sub-habits from your screenshot.",
               "actions": [
                 {
                   "type": "CreateHabit",

src/Orbit.Infrastructure/Services/Prompts/Sections/Static/ClarificationGuidanceSection.cs

@@ -13,20 +13,20 @@ public string Build(PromptContext context)
         sb.AppendLine("""
             ## Clarification Cards (NeedsClarification)
 
-            Some tools may return a structured "I need to ask the user a question" result instead of executing. This shows up as a `NeedsClarification` action containing a question + a small set of quick-action buttons (Daily, Weekly, X times per week, One-time task, etc.). The user taps one button and the tool re-runs with the chosen value merged into the original arguments — no follow-up tool call from you.
+            Some tools may return a structured "I need to ask the user a question" result instead of executing. This shows up as a `NeedsClarification` action containing a question + a small set of quick-action buttons (Daily, Weekly, X times per week, One-time task, etc.). The user taps one button and the tool re-runs with the chosen value merged into the original arguments, with no follow-up tool call from you.
 
             ### When this can happen
             - `create_habit` returns `NeedsClarification` if you call it with no `frequency_unit` AND the title contains "habit" / "rotina" / "hábito". Translation: you guessed it's a one-time task on a recurrence-sounding title. The tool refuses to silently create a one-time task in that case.
 
             ### How to behave when a tool returns NeedsClarification
-            - DO NOT add a plain-text question of your own on top — the card already asks. Replying with extra text duplicates the prompt and confuses the user.
-            - DO NOT immediately call another tool to "help out" — the user is now interacting with the card, not your text.
-            - A short acknowledgement like "Sure — what schedule would you like?" is OK but unnecessary; one short line max, or stay silent.
+            - DO NOT add a plain-text question of your own on top. The card already asks. Replying with extra text duplicates the prompt and confuses the user.
+            - DO NOT immediately call another tool to "help out". The user is now interacting with the card, not your text.
+            - A short acknowledgement like "Sure, what schedule would you like?" is OK but unnecessary; one short line max, or stay silent.
             - The user's button tap triggers the tool to re-run server-side. You will receive the resulting success/failure on the next turn just like any normal tool call.
 
             ### How to avoid the clarification in the first place
             - Prefer to ask the schedule INLINE in your text BEFORE calling `create_habit` when the user describes something as a "habit"/"rotina"/"hábito" without a schedule (see Structuring Strategy). The clarification card is the safety net; the ideal flow is for you to ask first.
-            - When the user clearly described a one-time task ("just once", "this Friday only", "uma vez"), call `create_habit` with no `frequency_unit` — the tool detects the explicit one-time language and does not request clarification.
+            - When the user clearly described a one-time task ("just once", "this Friday only", "uma vez"), call `create_habit` with no `frequency_unit`. The tool detects the explicit one-time language and does not request clarification.
 
             """);
         return sb.ToString();

src/Orbit.Infrastructure/Services/Prompts/Sections/Static/EncouragingToneSection.cs

@@ -13,8 +13,10 @@ public string Build(PromptContext context)
         sb.AppendLine("""
             ## Tone and Encouragement
 
-            Keep your voice warm, supportive, and concise. Celebrate progress and streaks with genuine, specific recognition, and give a finished week or a recovered streak a quick word of acknowledgement. Stay non-judgmental about missed days, treat a slip as a normal part of building habits, and point the way back without guilt or pressure. Be encouraging without being saccharine, skip the empty hype and exclamation spam, and just be a steady presence that helps the user keep moving.
+            Keep your voice warm, supportive, and concise. Write calmly and give progress and streaks a brief, specific acknowledgement. Stay non-judgmental about missed days, treat a slip as a normal part of building habits, and point the way back without guilt or pressure. Be encouraging without being saccharine, avoid empty hype, and be a steady presence that helps the user keep moving.
             """);
+        sb.AppendLine(NotificationVoice.CharacterRules);
+        sb.AppendLine("These bans apply to your prose. You may still set habit emoji fields through tools, and use markdown and bullet points in chat replies.");
         return sb.ToString();
     }
 }

tests/Orbit.Infrastructure.Tests/Services/NotificationVoiceTests.cs

@@ -37,6 +37,7 @@ public class NotificationVoiceTests
         { "daily summary, system", AiSummaryService.SystemPrompt },
         { "daily summary, en", DailySummaryPrompt("en") },
         { "daily summary, pt-BR", DailySummaryPrompt("pt-BR") },
+        { "chat, system with image", ChatSystemPrompt() },
     };
 
     public static TheoryData<string, string> EveryGeneratedUserPrompt() => new()
@@ -68,6 +69,35 @@ private static string DailySummaryPrompt(string language)
             new DailySummaryContext(today, today, today, language, null, 0, 0, new Dictionary<Guid, DateOnly>()));
     }
 
+    private static string ChatSystemPrompt()
+    {
+        var builder = new SystemPromptBuilder();
+        var request = new PromptBuildRequest(
+            [], [], HasImage: true, UserToday: new DateOnly(2026, 9, 18));
+
+        return builder.BuildStatic(request) + builder.BuildDynamic(request);
+    }
+
+    [Theory]
+    [InlineData("no exclamation mark")]
+    [InlineData("no emoji")]
+    [InlineData("no em dash")]
+    [InlineData("no en dash")]
+    [InlineData("no doubled hyphen")]
+    public void ChatPromptBansCharactersInEveryLanguageAndPreservesFormattingAndHabitEmojiTools(string ban)
+    {
+        var prompt = ChatSystemPrompt();
+
+        prompt.Should().Contain(ban);
+        prompt.Should().Contain("absolute in every language");
+        prompt.Should().Contain("reply's own words");
+        prompt.Should().Contain("habit emoji fields through tools");
+        prompt.Should().Contain("Use bullet points with newlines");
+        prompt.Should().Contain("bulk_update_habit_emojis");
+        prompt.Should().NotContain("no markdown");
+        prompt.Should().NotContain("no bullet");
+    }
+
     [Theory]
     [MemberData(nameof(EveryPromptSentToAModel))]
     public void NoPromptDemonstratesACharacterItBans(string label, string prompt)

tests/Orbit.Infrastructure.Tests/Services/PromptSectionTests.cs

@@ -90,6 +90,18 @@ public void Build_ContainsWarmEncouragingTone()
         result.Should().Contain("saccharine");
     }
 
+    [Fact]
+    public void Build_KeepsAcknowledgementCalmAndSpecific()
+    {
+        var ctx = new PromptContext(new List<Habit>(), new List<UserFact>(), false, null, null, null, null);
+        var result = new EncouragingToneSection().Build(ctx);
+
+        result.Should().Contain("calmly");
+        result.Should().Contain("specific acknowledgement");
+        result.Should().NotContain("Celebrate");
+        result.Should().NotContain("exclamation spam");
+    }
+
     [Fact]
     public void Build_ContainsNoEmOrEnDashes()
     {