Check if this issue is valid — if so, understand the root cause and fix it. At posthog/temporal/ai/session_summary/activities/a5_embed_and_store_segments.py, line 18:

<comment>Parameter `asset_id` is declared but never used in the function body. This is either dead code that should be removed, or indicates missing functionality where `asset_id` should be included in the metadata or used elsewhere.</comment>

<file context>
@@ -0,0 +1,77 @@
+async def embed_and_store_segments_activity(
+    inputs: VideoSummarySingleSessionInputs,
+    segments: list[VideoSegmentOutput],
+    asset_id: int,
+) -&gt; None:
+    &quot;&quot;&quot;Generate embeddings for all segments and produce to Kafka for ClickHouse storage
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At posthog/temporal/ai/session_summary/activities/a2_upload_video_to_gemini.py, line 48:

<comment>Missing `tmp_file.flush()` after writing to temp file. Since the upload reads the file by path (not file handle), the data must be flushed to disk first, otherwise it may upload an incomplete or empty file.</comment>

<file context>
@@ -0,0 +1,95 @@
+
+        # Write video to temporary file for upload
+        with tempfile.NamedTemporaryFile() as tmp_file:
+            tmp_file.write(video_bytes)
+            # Upload to Gemini
+            logger.info(
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At ee/hogai/videos/utils.py, line 16:

<comment>`track.duration` can be `None` for some media files, which would cause a `TypeError` when dividing by 1000.0. Add a null check before using the duration value.</comment>

<file context>
@@ -0,0 +1,17 @@
+    for track in media_info.tracks:
+        if track.track_type == &quot;General&quot;:
+            # Convert ms to seconds, ceil to avoid grey &quot;not-rendered&quot; frames at the start
+            return int(math.ceil(track.duration / 1000.0))
+    raise ValueError(&quot;No General track found in video to extract duration from&quot;)
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At posthog/temporal/ai/session_summary/activities/a1_export_session_video.py, line 105:

<comment>Using `uuid.uuid4()` in the workflow ID makes it non-deterministic across activity retries. If this activity is retried (e.g., due to timeout), a new workflow will be spawned instead of resuming the existing one. Use a deterministic ID based on `exported_asset.id` instead.</comment>

<file context>
@@ -0,0 +1,124 @@
+        await client.execute_workflow(
+            VideoExportWorkflow.run,
+            VideoExportInputs(exported_asset_id=exported_asset.id),
+            id=f&quot;session-video-summary-export_{inputs.session_id}_{uuid.uuid4()}&quot;,
+            task_queue=settings.VIDEO_EXPORT_TASK_QUEUE,
+            retry_policy=RetryPolicy(maximum_attempts=int(TEMPORAL_WORKFLOW_MAX_ATTEMPTS)),
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At posthog/temporal/ai/session_summary/activities/a3_analyze_video_segment.py, line 279:

<comment>`get_column_index` raises `ValueError` for missing columns, but this code expects it to return `None`. Sessions without exception events (which lack `$exception_types`, `$exception_values` columns) will cause this activity to crash. Wrap the call in try/except to handle missing columns gracefully.</comment>

<file context>
@@ -0,0 +1,328 @@
+    # Get indices for key columns
+    column_indices: dict[str, int | None] = {}
+    for col in key_columns:
+        column_indices[col] = get_column_index(simplified_events_columns, col)
+
+    # Format events
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At posthog/temporal/ai/session_summary/activities/a1_export_session_video.py, line 80:

<comment>Missing string concatenation: The second f-string is a standalone expression and won&#39;t be concatenated to `msg`. The error message will be incomplete, missing the duration details. Wrap both strings in parentheses.</comment>

<file context>
@@ -0,0 +1,130 @@
+
+        # Check if session is too short for summarization - note: this is different from the video duration, but probs close enough
+        if session_duration * 1000 &lt; MIN_SESSION_DURATION_FOR_SUMMARY_MS:
+            msg = f&quot;Session {inputs.session_id} video is too short for summarization &quot;
+            f&quot;({session_duration * 1000:.0f}ms &lt; {MIN_SESSION_DURATION_FOR_SUMMARY_MS}ms)&quot;
+            logger.error(msg, session_id=inputs.session_id, signals_type=&quot;session-summaries&quot;)
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At ee/hogai/videos/utils.py, line 16:

<comment>`track.duration` can be `None` in pymediainfo when duration metadata is unavailable, causing a `TypeError` on division. Consider checking for `None` before using the value.</comment>

<file context>
@@ -0,0 +1,17 @@
+    for track in media_info.tracks:
+        if track.track_type == &quot;General&quot;:
+            # Convert ms to seconds, ceil to avoid grey &quot;not-rendered&quot; frames at the start
+            return int(math.ceil(track.duration / 1000.0))
+    raise ValueError(&quot;No General track found in video to extract duration from&quot;)
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At posthog/temporal/ai/session_summary/activities/a4_consolidate_video_segments.py, line 95:

<comment>If the LLM returns `{&quot;segments&quot;: null}`, `parsed.get(&quot;segments&quot;, [])` returns `None` (not the default), causing a `TypeError` when iterating. Use `or []` to handle both missing keys and null values.</comment>

<file context>
@@ -0,0 +1,189 @@
+                confusion_detected=item.get(&quot;confusion_detected&quot;, False),
+                abandonment_detected=item.get(&quot;abandonment_detected&quot;, False),
+            )
+            for item in parsed.get(&quot;segments&quot;, [])
+        ]
+
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At posthog/temporal/ai/session_summary/activities/a2_upload_video_to_gemini.py, line 61:

<comment>Polling loop has no timeout mechanism. If Gemini&#39;s file processing hangs indefinitely, this activity will block forever. Consider adding a maximum wait time or iteration count with a timeout error.</comment>

<file context>
@@ -0,0 +1,95 @@
+                file=tmp_file.name, config=types.UploadFileConfig(mime_type=asset.export_format)
+            )
+            # Wait for file to be ready
+            while uploaded_file.state and uploaded_file.state.name == &quot;PROCESSING&quot;:
+                await asyncio.sleep(0.5)
+                logger.info(
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At posthog/temporal/ai/session_summary/activities/a2_upload_video_to_gemini.py, line 48:

<comment>Missing `tmp_file.flush()` after writing video bytes. When passing a temp file&#39;s name to another library for reading, you must flush the buffer first to ensure all data is written to disk. Without this, the Gemini upload may receive an empty or partial file.</comment>

<file context>
@@ -0,0 +1,95 @@
+
+        # Write video to temporary file for upload
+        with tempfile.NamedTemporaryFile() as tmp_file:
+            tmp_file.write(video_bytes)
+            # Upload to Gemini
+            logger.info(
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At posthog/temporal/ai/session_summary/activities/a3_analyze_video_segment.py, line 279:

<comment>`get_column_index` raises `ValueError` when a column is not found, but this code expects it to return `None` for missing columns. Optional columns like `$exception_types`, `$exception_values` may not exist in all event schemas, causing this to crash.</comment>

<file context>
@@ -0,0 +1,329 @@
+    # Get indices for key columns
+    column_indices: dict[str, int | None] = {}
+    for col in key_columns:
+        column_indices[col] = get_column_index(simplified_events_columns, col)
+
+    # Format events
</file context>