feat(insights): wire previous-analysis memory into the five remaining generators (#130)

Extends the v1.4 AI insights memory pattern established in PR #129
(general-status, blood-pressure-status) to the rest of the insight
surfaces. Each generator now reads its previous cached narrative via
`getPreviousInsightContext()` and injects the formatted block into the
user prompt so the model can call out improvements / regressions
relative to the user's last analysis instead of restating the status
from scratch every day.

Scopes wired in this commit:
- weight-status
- pulse-status
- bmi-status
- mood-status
- medication-compliance-status

No UI changes are needed — the insights card already renders the
comparison phrases the model emits. No schema change either; the
helper reads from the existing audit-log cache rows each generator
already writes to.

Co-authored-by: Marc-André Bombeck <noreply@anthropic.com>

src/lib/ai/prompts/bmi.ts

@@ -40,15 +40,20 @@ export function getBmiUserPrompt(
   snapshotJson: string,
   todayKey: string,
   locale: Locale,
+  previousContextBlock?: string,
 ): string {
+  const ctxBlock =
+    previousContextBlock && previousContextBlock.trim().length > 0
+      ? `\n\n${previousContextBlock}\n`
+      : "";
   if (locale === "en") {
     return `Date: ${todayKey} (Europe/Berlin)
-Analyse the BMI trajectory taking age, sex and weight trend into account. Place the WHO classification.
+Analyse the BMI trajectory taking age, sex and weight trend into account. Place the WHO classification.${ctxBlock}
 
 ${snapshotJson}`;
   }
   return `Datum: ${todayKey} (Europe/Berlin)
-Analysiere den BMI-Verlauf unter Berücksichtigung von Alter, Geschlecht und Gewichtstrend. Ordne die Klassifikation nach WHO ein.
+Analysiere den BMI-Verlauf unter Berücksichtigung von Alter, Geschlecht und Gewichtstrend. Ordne die Klassifikation nach WHO ein.${ctxBlock}
 
 ${snapshotJson}`;
 }

src/lib/ai/prompts/medication-compliance.ts

@@ -52,17 +52,22 @@ export function getMedicationComplianceUserPrompt(
   snapshotJson: string,
   todayKey: string,
   locale: Locale,
+  previousContextBlock?: string,
 ): string {
+  const ctxBlock =
+    previousContextBlock && previousContextBlock.trim().length > 0
+      ? `\n\n${previousContextBlock}\n`
+      : "";
   if (locale === "en") {
     return `Date: ${todayKey} (Europe/Berlin)
 Analyse medication adherence with focus on patterns, effectiveness correlation and concrete suggestions for improvement.
-Use the correlation data and historical comparison to back up the link between adherence and vital signs.
+Use the correlation data and historical comparison to back up the link between adherence and vital signs.${ctxBlock}
 
 ${snapshotJson}`;
   }
   return `Datum: ${todayKey} (Europe/Berlin)
 Analysiere die Medikamenten-Einnahmetreue mit Fokus auf Muster, Wirksamkeitskorrelation und konkrete Verbesserungsvorschläge.
-Nutze die Korrelationsdaten und den historischen Vergleich um den Zusammenhang zwischen Adhärenz und Vitalwerten zu belegen.
+Nutze die Korrelationsdaten und den historischen Vergleich um den Zusammenhang zwischen Adhärenz und Vitalwerten zu belegen.${ctxBlock}
 
 ${snapshotJson}`;
 }

src/lib/ai/prompts/mood.ts

@@ -30,15 +30,20 @@ export function getMoodUserPrompt(
   snapshotJson: string,
   todayKey: string,
   locale: Locale,
+  previousContextBlock?: string,
 ): string {
+  const ctxBlock =
+    previousContextBlock && previousContextBlock.trim().length > 0
+      ? `\n\n${previousContextBlock}\n`
+      : "";
   if (locale === "en") {
     return `Date: ${todayKey} (Europe/Berlin)
-Analyse the mood data with focus on trend, stability and links to other health metrics.
+Analyse the mood data with focus on trend, stability and links to other health metrics.${ctxBlock}
 
 ${snapshotJson}`;
   }
   return `Datum: ${todayKey} (Europe/Berlin)
-Analysiere die Stimmungsdaten mit Fokus auf Trend, Stabilitat und Zusammenhange mit anderen Gesundheitsmetriken.
+Analysiere die Stimmungsdaten mit Fokus auf Trend, Stabilitat und Zusammenhange mit anderen Gesundheitsmetriken.${ctxBlock}
 
 ${snapshotJson}`;
 }

src/lib/ai/prompts/pulse.ts

@@ -58,17 +58,22 @@ export function getPulseUserPrompt(
   snapshotJson: string,
   todayKey: string,
   locale: Locale,
+  previousContextBlock?: string,
 ): string {
+  const ctxBlock =
+    previousContextBlock && previousContextBlock.trim().length > 0
+      ? `\n\n${previousContextBlock}\n`
+      : "";
   if (locale === "en") {
     return `Date: ${todayKey} (Europe/Berlin)
 Analyse the pulse / heart-rate data with focus on resting-pulse trend, variability and links to medication and mood.
-Use the precomputed correlations and historical comparison for a sound analysis.
+Use the precomputed correlations and historical comparison for a sound analysis.${ctxBlock}
 
 ${snapshotJson}`;
   }
   return `Datum: ${todayKey} (Europe/Berlin)
 Analysiere die Puls-/Herzfrequenzdaten mit Fokus auf Ruhepulstrend, Variabilität und Zusammenhänge mit Medikation und Stimmung.
-Nutze die vorberechneten Korrelationen und den historischen Vergleich für eine fundierte Analyse.
+Nutze die vorberechneten Korrelationen und den historischen Vergleich für eine fundierte Analyse.${ctxBlock}
 
 ${snapshotJson}`;
 }

src/lib/ai/prompts/weight.ts

@@ -52,17 +52,22 @@ export function getWeightUserPrompt(
   snapshotJson: string,
   todayKey: string,
   locale: Locale,
+  previousContextBlock?: string,
 ): string {
+  const ctxBlock =
+    previousContextBlock && previousContextBlock.trim().length > 0
+      ? `\n\n${previousContextBlock}\n`
+      : "";
   if (locale === "en") {
     return `Date: ${todayKey} (Europe/Berlin)
 Analyse the weight trajectory with focus on trends, BMI classification and links to other vital signs.
-Use the temporal layers (avg7, avg30, avg90, allTime) and the historical comparison for a nuanced assessment.
+Use the temporal layers (avg7, avg30, avg90, allTime) and the historical comparison for a nuanced assessment.${ctxBlock}
 
 ${snapshotJson}`;
   }
   return `Datum: ${todayKey} (Europe/Berlin)
 Analysiere die Gewichtsentwicklung mit Fokus auf Trends, BMI-Klassifikation und Zusammenhang mit anderen Vitalwerten.
-Nutze die temporalen Schichten (avg7, avg30, avg90, allTime) und den historischen Vergleich für eine differenzierte Bewertung.
+Nutze die temporalen Schichten (avg7, avg30, avg90, allTime) und den historischen Vergleich für eine differenzierte Bewertung.${ctxBlock}
 
 ${snapshotJson}`;
 }

src/lib/insights/bmi-status.ts

@@ -3,6 +3,10 @@ import { resolveProvider } from "@/lib/ai/provider";
 import { getBmiSystemPrompt, getBmiUserPrompt } from "@/lib/ai/prompts/bmi";
 import { classifyBMI } from "@/lib/analytics/classifications";
 import { getNoKeyBmiStatusText } from "@/lib/insights/no-key-fallbacks";
+import {
+  formatPreviousContextForPrompt,
+  getPreviousInsightContext,
+} from "@/lib/insights/memory";
 
 const BMI_STATUS_POINTS = 30;
 
@@ -238,9 +242,25 @@ export async function generateBmiStatusForUser(
 
   const snapshotJson = JSON.stringify(snapshot, null, 2);
 
+  const previousContext = await getPreviousInsightContext(
+    userId,
+    "bmi-status",
+    locale,
+    12,
+  );
+  const previousContextBlock = formatPreviousContextForPrompt(
+    previousContext,
+    locale,
+  );
+
   const result = await provider.generateCompletion({
     systemPrompt: getBmiSystemPrompt(locale),
-    userPrompt: getBmiUserPrompt(snapshotJson, todayKey, locale),
+    userPrompt: getBmiUserPrompt(
+      snapshotJson,
+      todayKey,
+      locale,
+      previousContextBlock,
+    ),
     temperature: 0.3,
     maxTokens: 1000,
   });

src/lib/insights/medication-compliance-status.ts

@@ -1,10 +1,17 @@
 import { prisma } from "@/lib/db";
 import { resolveProvider } from "@/lib/ai/provider";
-import { getMedicationComplianceSystemPrompt, getMedicationComplianceUserPrompt } from "@/lib/ai/prompts/medication-compliance";
+import {
+  getMedicationComplianceSystemPrompt,
+  getMedicationComplianceUserPrompt,
+} from "@/lib/ai/prompts/medication-compliance";
 import { calculateCompliance } from "@/lib/analytics/compliance";
 import { getMedicationCategories } from "@/lib/medication-category";
 import { sanitizeForPrompt } from "@/lib/insights/sanitize";
 import { getNoKeyMedicationComplianceStatusText } from "@/lib/insights/no-key-fallbacks";
+import {
+  formatPreviousContextForPrompt,
+  getPreviousInsightContext,
+} from "@/lib/insights/memory";
 
 const MEDICATION_COMPLIANCE_STATUS_POINTS = 30;
 
@@ -291,9 +298,25 @@ export async function generateMedicationComplianceStatusForUser(
 
   const snapshotJson = JSON.stringify(snapshot, null, 2);
 
+  const previousContext = await getPreviousInsightContext(
+    userId,
+    "medication-compliance-status",
+    locale,
+    12,
+  );
+  const previousContextBlock = formatPreviousContextForPrompt(
+    previousContext,
+    locale,
+  );
+
   const result = await provider.generateCompletion({
     systemPrompt: getMedicationComplianceSystemPrompt(locale),
-    userPrompt: getMedicationComplianceUserPrompt(snapshotJson, todayKey, locale),
+    userPrompt: getMedicationComplianceUserPrompt(
+      snapshotJson,
+      todayKey,
+      locale,
+      previousContextBlock,
+    ),
     temperature: 0.3,
     maxTokens: 1000,
   });

src/lib/insights/mood-status.ts

@@ -2,6 +2,10 @@ import { prisma } from "@/lib/db";
 import { resolveProvider } from "@/lib/ai/provider";
 import { getMoodSystemPrompt, getMoodUserPrompt } from "@/lib/ai/prompts/mood";
 import { getNoKeyMoodStatusText } from "@/lib/insights/no-key-fallbacks";
+import {
+  formatPreviousContextForPrompt,
+  getPreviousInsightContext,
+} from "@/lib/insights/memory";
 import {
   pearsonCorrelation,
   type PairedPoint,
@@ -212,8 +216,7 @@ export async function generateMoodStatusForUser(
       ? null
       : round(
           (moodSeries.filter(
-            (entry) =>
-              entry.value >= greenMin && entry.value <= greenMax,
+            (entry) => entry.value >= greenMin && entry.value <= greenMax,
           ).length /
             moodSeries.length) *
             100,
@@ -226,9 +229,7 @@ export async function generateMoodStatusForUser(
 
   const oldestEntry = entries.length > 0 ? entries[0].moodLoggedAt : null;
   const newestEntry =
-    entries.length > 0
-      ? entries[entries.length - 1].moodLoggedAt
-      : null;
+    entries.length > 0 ? entries[entries.length - 1].moodLoggedAt : null;
   const totalSpanDays =
     oldestEntry && newestEntry
       ? Math.round(
@@ -237,9 +238,7 @@ export async function generateMoodStatusForUser(
         )
       : 0;
   const newestEntryDaysAgo = newestEntry
-    ? Math.round(
-        (Date.now() - newestEntry.getTime()) / (24 * 60 * 60 * 1000),
-      )
+    ? Math.round((Date.now() - newestEntry.getTime()) / (24 * 60 * 60 * 1000))
     : null;
 
   // Fetch cross-metric context for enrichment
@@ -335,7 +334,9 @@ export async function generateMoodStatusForUser(
       tags: tagSummary.length > 0 ? tagSummary : null,
     },
     crossMetricContext:
-      weightSeries.length >= 3 || sysSeries.length >= 3 || pulseSeries.length >= 3
+      weightSeries.length >= 3 ||
+      sysSeries.length >= 3 ||
+      pulseSeries.length >= 3
         ? {
             weight:
               weightSeries.length >= 3
@@ -364,9 +365,25 @@ export async function generateMoodStatusForUser(
 
   const snapshotJson = JSON.stringify(snapshot, null, 2);
 
+  const previousContext = await getPreviousInsightContext(
+    userId,
+    "mood-status",
+    locale,
+    12,
+  );
+  const previousContextBlock = formatPreviousContextForPrompt(
+    previousContext,
+    locale,
+  );
+
   const result = await provider.generateCompletion({
     systemPrompt: getMoodSystemPrompt(locale),
-    userPrompt: getMoodUserPrompt(snapshotJson, todayKey, locale),
+    userPrompt: getMoodUserPrompt(
+      snapshotJson,
+      todayKey,
+      locale,
+      previousContextBlock,
+    ),
     temperature: 0.3,
     maxTokens: 1000,
   });

src/lib/insights/pulse-status.ts

@@ -1,6 +1,13 @@
 import { prisma } from "@/lib/db";
 import { resolveProvider } from "@/lib/ai/provider";
-import { getPulseSystemPrompt, getPulseUserPrompt } from "@/lib/ai/prompts/pulse";
+import {
+  getPulseSystemPrompt,
+  getPulseUserPrompt,
+} from "@/lib/ai/prompts/pulse";
+import {
+  formatPreviousContextForPrompt,
+  getPreviousInsightContext,
+} from "@/lib/insights/memory";
 import {
   getAgeFromDateOfBirth,
   getPersonalizedPulseTarget,
@@ -286,9 +293,25 @@ export async function generatePulseStatusForUser(
 
   const snapshotJson = JSON.stringify(snapshot, null, 2);
 
+  const previousContext = await getPreviousInsightContext(
+    userId,
+    "pulse-status",
+    locale,
+    12,
+  );
+  const previousContextBlock = formatPreviousContextForPrompt(
+    previousContext,
+    locale,
+  );
+
   const result = await provider.generateCompletion({
     systemPrompt: getPulseSystemPrompt(locale),
-    userPrompt: getPulseUserPrompt(snapshotJson, todayKey, locale),
+    userPrompt: getPulseUserPrompt(
+      snapshotJson,
+      todayKey,
+      locale,
+      previousContextBlock,
+    ),
     temperature: 0.3,
     maxTokens: 1000,
   });

src/lib/insights/weight-status.ts

@@ -1,11 +1,18 @@
 import { prisma } from "@/lib/db";
 import { resolveProvider } from "@/lib/ai/provider";
-import { getWeightSystemPrompt, getWeightUserPrompt } from "@/lib/ai/prompts/weight";
+import {
+  getWeightSystemPrompt,
+  getWeightUserPrompt,
+} from "@/lib/ai/prompts/weight";
 import {
   pearsonCorrelation,
   type PairedPoint,
 } from "@/lib/analytics/correlations";
 import { getNoKeyWeightStatusText } from "@/lib/insights/no-key-fallbacks";
+import {
+  formatPreviousContextForPrompt,
+  getPreviousInsightContext,
+} from "@/lib/insights/memory";
 
 const WEIGHT_STATUS_POINTS = 30;
 
@@ -341,7 +348,10 @@ export async function generateWeightStatusForUser(
             latest: dailyMoodSeries.at(-1)?.value ?? null,
             series: dailyMoodSeries.slice(-10),
             moodVsWeightCorrelation: (() => {
-              const moodVsWeightPairs = pairDailySeries(dailyMoodSeries, weightSeries);
+              const moodVsWeightPairs = pairDailySeries(
+                dailyMoodSeries,
+                weightSeries,
+              );
               return pearsonCorrelation(moodVsWeightPairs);
             })(),
           }
@@ -350,9 +360,25 @@ export async function generateWeightStatusForUser(
 
   const snapshotJson = JSON.stringify(snapshot, null, 2);
 
+  const previousContext = await getPreviousInsightContext(
+    userId,
+    "weight-status",
+    locale,
+    12,
+  );
+  const previousContextBlock = formatPreviousContextForPrompt(
+    previousContext,
+    locale,
+  );
+
   const result = await provider.generateCompletion({
     systemPrompt: getWeightSystemPrompt(locale),
-    userPrompt: getWeightUserPrompt(snapshotJson, todayKey, locale),
+    userPrompt: getWeightUserPrompt(
+      snapshotJson,
+      todayKey,
+      locale,
+      previousContextBlock,
+    ),
     temperature: 0.3,
     maxTokens: 1000,
   });