apps/web-app/.env.example (1)

27-27: Add brief guidance for the new prompt var.

Consider a one‑line comment above with expected format (tone, JSON schema, language) to reduce misconfiguration.

apps/web-app/server/tasks/ai/weekly-report.ts (3)

26-51: Avoid O(N²) counting and repeat lookups; precompute maps.

preparePerformer() filters tasks on every call. Build staffById and completedCounts once, then reuse.

Apply this diff:

-      function preparePerformer(performerId: string | null) {
-        const user = staff.find((user) => user.id === performerId)
-        if (!user) {
-          return null
-        }
-
-        return {
-          name: user?.name,
-          surname: user?.surname,
-          caption: user?.caption,
-          completedTasksCount: tasks.filter((task) => task.performerId === user?.id).length,
-        }
-      }
+      const staffById = new Map(staff.map(u => [u.id, u]))
+      const completedCounts = new Map<string, number>()
+      for (const t of tasks) {
+        if (!t.performerId) continue
+        completedCounts.set(t.performerId, (completedCounts.get(t.performerId) ?? 0) + 1)
+      }
+
+      function preparePerformer(performerId: string | null) {
+        if (!performerId) return null
+        const user = staffById.get(performerId)
+        if (!user) return null
+        return {
+          name: user.name,
+          surname: user.surname,
+          caption: user.caption,
+          completedTasksCount: completedCounts.get(user.id) ?? 0,
+        }
+      }

---

71-74: Defensive read of completion result + log empty.

Guard against empty choices and add a warning for observability.

Apply this diff:

-      const finalMessage = response.choices[0].message.content
-      if (!finalMessage) {
-        return { result: true }
-      }
+      const finalMessage = response.choices?.[0]?.message?.content?.trim()
+      if (!finalMessage) {
+        logger.warn('ai:weekly-report returned empty content')
+        return { result: true }
+      }

---

83-88: Error handling policy: success on failure hides issues.

Catching and then returning success can mask outages. Consider returning { result: false } or emitting an alert/metric.

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

apps/web-app/server/tasks/ai/weekly-report.ts (5)

23-25: Skip OpenAI call when there’s nothing to summarize.

Avoid paid call and flow write on empty weeks.

Apply:

       const tasks = await repository.task.listCompletedThisWeek()
       const staff = await repository.user.findStaff()

+      if (tasks.length === 0) {
+        logger.info('No completed tasks this week; skipping AI call and flow item creation')
+        return { result: true }
+      }

---

26-38: Avoid O(n²) scans per performer; preindex staff and counts.

Precompute maps; current filter-per-task per user is quadratic.

Apply within this function:

-      function preparePerformer(performerId: string | null) {
-        const user = staff.find((user) => user.id === performerId)
-        if (!user) {
-          return null
-        }
-
-        return {
-          name: user?.name,
-          surname: user?.surname,
-          caption: user?.caption,
-          completedTasksCount: tasks.filter((task) => task.performerId === user?.id).length,
-        }
-      }
+      function preparePerformer(performerId: string | null) {
+        if (!performerId) return null
+        const user = staffMap.get(performerId)
+        if (!user) return null
+        return {
+          name: user.name,
+          surname: user.surname,
+          caption: user.caption,
+          completedTasksCount: completedCounts.get(performerId) ?? 0,
+        }
+      }

Add just above this function:

const staffMap = new Map(staff.map(u => [u.id, u]))
const completedCounts = tasks.reduce((acc, t) => {
  if (t.performerId) acc.set(t.performerId, (acc.get(t.performerId) ?? 0) + 1)
  return acc
}, new Map<string, number>())

---

59-71: Bound cost and increase determinism.

Set low temperature and cap tokens; also tag requests with user.

Apply:

       const response = await client.chat.completions.create({
         model: ai.modelPro,
+        user: 'ai:weekly-report',
+        temperature: 0.2,
+        max_tokens: 800,
         messages: [

If the weekly payload can grow, consider chunking tasks and asking the model to merge summaries.

---

73-76: Defensive parse of completion and finish reason.

Avoid empty/unfinished responses.

Apply:

-      const finalMessage = response.choices[0].message.content
-      if (!finalMessage) {
+      const choice = response.choices[0]
+      const finalMessage = choice.message.content?.trim()
+      if (!finalMessage || choice.finish_reason !== 'stop') {
+        logger.warn('OpenAI response missing content or not finished: %o', { finish_reason: choice.finish_reason })
         return { result: true }
       }

---

79-84: Use ISO/padded week for stable titles.

'w' is locale‑week and unpadded. Prefer ISO ('II') or at least padded locale ('ww').

Apply one:

-      const week = format(new Date(), 'w', { locale: ru })
+      const week = format(new Date(), 'II') // ISO week 01–53

apps/web-app/server/tasks/ai/daily-report.ts (2)

23-24: Short‑circuit on empty day.

Skip OpenAI call when no completed tasks.

Apply:

       const tasks = await repository.task.listCompletedToday()
+      if (tasks.length === 0) {
+        logger.info('No completed tasks today; skipping AI call and flow item creation')
+        return { result: true }
+      }

---

38-50: Bound cost and increase determinism.

Mirror weekly: set user, low temperature, and max_tokens.

Apply:

       const response = await client.chat.completions.create({
         model: ai.modelPro,
+        user: 'ai:daily-report',
+        temperature: 0.2,
+        max_tokens: 600,
         messages: [

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro