feat(ai): surface server-dropped agent-activity records

The server reports how many records it could not scan and dropped (never
stored). Thread that count through AgentActivityBatchResult and the report, and
print it in the `ai discover --history` summary when non-zero, so drops are
visible instead of silently making the counts not add up.

Issue: NHI-1628

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: amascia-gg

ggshield/cmd/ai/discover.py

@@ -148,6 +148,7 @@ def _summarize_discovery(
             "ingested": activity_report.ingested,
             "duplicates": activity_report.duplicates,
             "failed_batches": activity_report.failed_batches,
+            "dropped": activity_report.dropped,
         }
     return summary
 
@@ -223,3 +224,6 @@ def print_summary(summary: Dict[str, Any]) -> None:
         if agent_activity.get("failed_batches", 0) > 0:
             label = format_text("Failed batches:", STYLE["detector_line_start"])
             click.echo(f"  • {label} {agent_activity['failed_batches']:,}")
+        if agent_activity.get("dropped", 0) > 0:
+            label = format_text("Dropped (unscannable):", STYLE["detector_line_start"])
+            click.echo(f"  • {label} {agent_activity['dropped']:,}")

ggshield/verticals/ai/agent_activity/orchestrator.py

@@ -29,6 +29,7 @@ class AgentActivityBatchResult:
     ingested: int
     duplicates: int
     success: bool
+    dropped: int = 0
 
 
 @dataclass
@@ -37,14 +38,14 @@ class AgentActivityReport:
     ingested: int = 0
     duplicates: int = 0
     failed_batches: int = 0
+    # Records the server could not scan and dropped (never stored).
+    dropped: int = 0
 
 
 def send_agent_activity_batch(
     client: GGClient, events: List[AgentActivityEvent], user: UserInfo
 ) -> AgentActivityBatchResult:
     """Serialise events and submit them as one batch (with the reporting user)."""
-    if not events:
-        return AgentActivityBatchResult(ingested=0, duplicates=0, success=True)
     payload = [e.to_dict() for e in events]
     response = client.send_agent_activity(payload, user=user.to_dict())
     if isinstance(response, Detail):
@@ -54,6 +55,7 @@ def send_agent_activity_batch(
         ingested=response.ingested,
         duplicates=response.duplicates,
         success=True,
+        dropped=getattr(response, "dropped", 0),
     )
 
 
@@ -83,6 +85,7 @@ def flush() -> None:
         else:
             report.ingested += result.ingested
             report.duplicates += result.duplicates
+            report.dropped += result.dropped
             if not result.success:
                 report.failed_batches += 1
         buffer.clear()

tests/unit/verticals/ai/agent_activity/test_orchestrator.py

@@ -18,7 +18,7 @@
 def test_send_agent_activity_batch_serialises_and_calls_client() -> None:
     client = MagicMock()
     client.send_agent_activity.return_value = MagicMock(
-        success=True, ingested=2, duplicates=0
+        success=True, ingested=2, duplicates=0, dropped=0
     )
     events = [
         AgentActivityEvent(
@@ -59,13 +59,6 @@ def test_send_agent_activity_batch_serialises_and_calls_client() -> None:
     assert result.ingested == 2
 
 
-def test_send_agent_activity_batch_empty_list_short_circuits() -> None:
-    client = MagicMock()
-    result = send_agent_activity_batch(client, [], _USER)
-    client.send_agent_activity.assert_not_called()
-    assert result.ingested == 0
-
-
 def test_send_agent_activity_batch_treats_detail_as_failure() -> None:
     """An API error (Detail) is reported as a failed batch, not an ingest."""
     client = MagicMock()
@@ -105,7 +98,7 @@ def _event(agent_name: str, i: int) -> AgentActivityEvent:
 def test_collect_agent_activity_attaches_user_to_each_batch() -> None:
     client = MagicMock()
     client.send_agent_activity.return_value = MagicMock(
-        success=True, ingested=1, duplicates=0
+        success=True, ingested=1, duplicates=0, dropped=0
     )
     collect_agent_activity(client, _USER, [_make_agent("a", [_event("a", 1)])])
 
@@ -120,7 +113,7 @@ def test_collect_agent_activity_batches_in_chunks_of_batch_size() -> None:
     agent = _make_agent("a", events)
     client = MagicMock()
     client.send_agent_activity.return_value = MagicMock(
-        success=True, ingested=BATCH_SIZE, duplicates=0
+        success=True, ingested=BATCH_SIZE, duplicates=0, dropped=0
     )
 
     report = collect_agent_activity(client, _USER, [agent])
@@ -134,7 +127,7 @@ def test_collect_agent_activity_aggregates_per_agent_counts() -> None:
     events_b = [_event("b", 2)]
     client = MagicMock()
     client.send_agent_activity.return_value = MagicMock(
-        success=True, ingested=2, duplicates=0
+        success=True, ingested=2, duplicates=0, dropped=0
     )
 
     report = collect_agent_activity(
@@ -166,7 +159,9 @@ def test_collect_agent_activity_flushes_on_byte_threshold() -> None:
         for i in range(3)
     ]
     client = MagicMock()
-    client.send_agent_activity.return_value = MagicMock(ingested=1, duplicates=0)
+    client.send_agent_activity.return_value = MagicMock(
+        ingested=1, duplicates=0, dropped=0
+    )
 
     report = collect_agent_activity(client, _USER, [_make_agent("a", events)])
 
@@ -184,3 +179,14 @@ def test_collect_agent_activity_counts_network_error_as_failed_batch() -> None:
     assert report.parsed == 1
     assert report.ingested == 0
     assert report.failed_batches == 1
+
+
+def test_collect_agent_activity_aggregates_dropped_from_response() -> None:
+    """The server's dropped count (records it could not scan) is surfaced."""
+    client = MagicMock()
+    client.send_agent_activity.return_value = MagicMock(
+        success=True, ingested=0, duplicates=0, dropped=3
+    )
+    report = collect_agent_activity(client, _USER, [_make_agent("a", [_event("a", 1)])])
+
+    assert report.dropped == 3