handoff fixes

backend/app/main.py

@@ -194,6 +194,20 @@ async def lifespan(_: FastAPI):
         [("message_id", 1)],
         sparse=True,
     )
+    await mongo_db.messages.create_index(
+        [("message_id_folded", 1)],
+        sparse=True,
+    )
+    await mongo_db.messages.create_index(
+        [("generated_from_handoff_message_id", 1)],
+        sparse=True,
+        unique=True,
+    )
+    await mongo_db.messages.create_index(
+        [("mirrored_from_external_message_id", 1)],
+        sparse=True,
+        unique=True,
+    )
     await mongo_db.mailbox_entries.create_index(
         [("account_id", 1), ("thread_id", 1), ("event_at", -1), ("_id", -1)]
     )

backend/app/routers/agent_handoff.py

@@ -18,6 +18,7 @@ from app.schemas import (
     HandoffTestRequest,
     HandoffTestResult,
 )
+from app.services.handoff_mirror_policy import is_relay_acknowledgment_message
 from app.services.human_handoff_service import MESSAGE_SCOPE_HANDOFF
 from app.services.handoff_evaluator import EMAIL_TYPE_META, HANDOFF_LABELS, evaluate_handoff
 
@@ -261,8 +262,15 @@ async def get_handoff_conversation(
     messages: list[HandoffMessageRead] = []
     for row in rows:
         message = row["message"]
+        if is_relay_acknowledgment_message(message):
+            continue
         sender_address = (message.get("sender") or {}).get("address", "").lower()
-        role = "human" if sender_address == human_email else "agent"
+        if message.get("mirrored_from_external_message_id"):
+            role = "customer"
+        elif sender_address == human_email:
+            role = "human"
+        else:
+            role = "agent"
         event_at = row.get("event_at") or message.get("received_at") or message.get("sent_at") or message["created_at"]
         raw_body = message.get("raw_body")
         new_reply_content = message.get("new_reply_content")
@@ -281,6 +289,7 @@ async def get_handoff_conversation(
                     and new_reply_content
                     and raw_body.strip() != new_reply_content.strip()
                 ),
+                relay_status=message.get("relay_status") if role == "human" else None,
                 event_at=event_at,
             )
         )

backend/app/routers/emails.py

@@ -24,7 +24,7 @@ from app.schemas import (
 )
 from app.services.domain_provisioning import get_domain_by_name
 from app.services.mail_sender import deliver_email
-from app.services.message_ids import normalize_message_id
+from app.services.message_ids import canonical_message_id, message_id_fields
 from app.services.mime_parser import parse_mime
 from app.services.raw_mime_storage import get_raw_mime_storage
 from app.services.reply_recipients import reply_recipients
@@ -249,9 +249,8 @@ async def create_email(
             references=threading_fields.get("references"),
         )
         document["status"] = "accepted"
-        normalized_id = normalize_message_id(internet_message_id)
-        document["message_id"] = normalized_id
-        document["internet_message_id"] = normalized_id
+        normalized_id = canonical_message_id(internet_message_id)
+        document.update(message_id_fields(normalized_id))
         log_smtp_message_sending(
             generated_message_id=normalized_id,
             in_reply_to=threading_fields.get("in_reply_to"),

backend/app/schemas.py

@@ -310,13 +310,14 @@ class HandoffConversationSummary(BaseModel):
 
 class HandoffMessageRead(BaseModel):
     id: str
-    role: Literal["agent", "human"]
+    role: Literal["agent", "human", "customer"]
     subject: str
     snippet: str
     body_text: str
     new_reply_content: str | None = None
     raw_body: str | None = None
     has_quoted_text: bool = False
+    relay_status: str | None = None
     event_at: datetime
 
 

backend/app/services/external_reply_service.py

@@ -7,17 +7,27 @@ from bson import ObjectId
 
 from app.models import MailDomain
 from app.services.handoff_email_templates import external_reply_subject
-from app.services.mail_sender import deliver_email
-from app.services.message_ids import normalize_message_id
-from app.services.thread_service import (
-    attach_message_to_thread,
-    message_event_at,
-    threading_fields_from_parent,
+from app.services.mail_sender import deliver_email, log_outbound_mime_headers
+from app.services.message_ids import (
+    build_references_header,
+    canonical_message_id,
+    folded_message_id,
+    first_message_id,
+    internet_message_id_of_message,
+    message_id_fields,
 )
+from app.services.mime_parser import parse_mime
+from app.services.raw_mime_storage import get_raw_mime_storage
+from app.services.thread_service import threading_fields_from_parent
 
 logger = logging.getLogger(__name__)
 
 MESSAGE_SCOPE_EXTERNAL = "EXTERNAL"
+MESSAGE_SCOPE_HANDOFF = "HANDOFF"
+
+
+class ExternalReplyThreadingError(RuntimeError):
+    pass
 
 
 @dataclass(frozen=True)
@@ -38,9 +48,169 @@ class ExternalReplyPlan:
 
 
 def internet_message_id_of(message: dict) -> str | None:
-    return normalize_message_id(
+    stored = canonical_message_id(
         message.get("internet_message_id") or message.get("message_id")
     )
+    raw_header = (message.get("raw_headers") or {}).get("message_id")
+    raw_canonical = canonical_message_id(raw_header)
+    if raw_canonical and stored:
+        if folded_message_id(raw_canonical) == folded_message_id(stored):
+            return raw_canonical
+    return stored or raw_canonical
+
+
+async def resolve_external_reply_parent(
+    database,
+    *,
+    account_id: str,
+    external_thread_id: ObjectId,
+    customer_address: str,
+    handoff: dict | None = None,
+) -> dict | None:
+    """Pick the latest external inbound customer message with a real Message-ID."""
+    customer = customer_address.lower()
+    pipeline = [
+        {
+            "$match": {
+                "account_id": account_id,
+                "thread_id": external_thread_id,
+                "message_scope": MESSAGE_SCOPE_EXTERNAL,
+                "deletion_state": {"$in": [None, ""]},
+            }
+        },
+        {"$sort": {"event_at": -1, "_id": -1}},
+        {
+            "$lookup": {
+                "from": "messages",
+                "localField": "message_id",
+                "foreignField": "_id",
+                "as": "message",
+            }
+        },
+        {"$unwind": "$message"},
+        {
+            "$match": {
+                "message.message_scope": {"$in": [MESSAGE_SCOPE_EXTERNAL, None]},
+                "message.direction": "inbound",
+                "$or": [
+                    {"message.internet_message_id": {"$gt": ""}},
+                    {"message.message_id": {"$gt": ""}},
+                ],
+            }
+        },
+        {"$limit": 1},
+    ]
+    cursor = await database.mailbox_entries.aggregate(pipeline)
+    rows = [row async for row in cursor]
+    if rows:
+        return rows[0]["message"]
+
+    if handoff and handoff.get("trigger_external_message_id"):
+        trigger = await database.messages.find_one(
+            {
+                "_id": handoff["trigger_external_message_id"],
+                "message_scope": MESSAGE_SCOPE_EXTERNAL,
+            }
+        )
+        if trigger and internet_message_id_of(trigger):
+            return trigger
+
+    trigger_internet = handoff.get("trigger_internet_message_id") if handoff else None
+    if trigger_internet:
+        folded = folded_message_id(trigger_internet)
+        trigger = await database.messages.find_one(
+            {
+                "message_scope": MESSAGE_SCOPE_EXTERNAL,
+                "message_id_folded": folded,
+            }
+        )
+        if trigger:
+            return trigger
+
+    return None
+
+
+def validate_external_reply_parent(parent_external_message: dict) -> None:
+    parent_id = internet_message_id_of(parent_external_message)
+    if not parent_id:
+        raise ExternalReplyThreadingError(
+            "External reply parent is missing internet_message_id"
+        )
+    scope = parent_external_message.get("message_scope")
+    if scope == MESSAGE_SCOPE_HANDOFF:
+        raise ExternalReplyThreadingError(
+            "External reply parent must not be a HANDOFF-scoped message"
+        )
+
+
+async def assert_in_reply_to_resolves_to_external_parent(
+    database,
+    in_reply_to: str | None,
+) -> dict:
+    """Ensure In-Reply-To points at an EXTERNAL message, never HANDOFF."""
+    canonical = canonical_message_id(in_reply_to)
+    folded = folded_message_id(canonical)
+    if not canonical or not folded:
+        raise ExternalReplyThreadingError("External reply is missing In-Reply-To")
+
+    parent = await database.messages.find_one(
+        {
+            "$or": [
+                {"message_id_folded": folded},
+                {"message_id": canonical},
+                {"internet_message_id": canonical},
+            ],
+        }
+    )
+    if parent is None:
+        raise ExternalReplyThreadingError(
+            f"In-Reply-To {canonical} does not resolve to a known message"
+        )
+    scope = parent.get("message_scope")
+    if scope == MESSAGE_SCOPE_HANDOFF:
+        raise ExternalReplyThreadingError(
+            f"In-Reply-To {canonical} resolves to HANDOFF-scoped message"
+        )
+    if scope not in (MESSAGE_SCOPE_EXTERNAL, None):
+        raise ExternalReplyThreadingError(
+            f"In-Reply-To {canonical} resolves to non-external message scope={scope}"
+        )
+    return parent
+
+
+async def assert_external_threading_chain(
+    database,
+    *,
+    in_reply_to: str | None,
+    references: list[str] | None,
+) -> None:
+    """Ensure outbound headers only reference EXTERNAL messages."""
+    await assert_in_reply_to_resolves_to_external_parent(database, in_reply_to)
+
+    folded_targets = {
+        folded_message_id(value)
+        for value in [*(references or [])]
+        if folded_message_id(value)
+    }
+    in_reply_folded = folded_message_id(in_reply_to)
+    if in_reply_folded:
+        folded_targets.discard(in_reply_folded)
+
+    for folded in folded_targets:
+        handoff_hit = await database.messages.find_one(
+            {
+                "$or": [
+                    {"message_id_folded": folded},
+                    {"message_id": folded},
+                ],
+                "message_scope": MESSAGE_SCOPE_HANDOFF,
+            },
+            {"_id": 1},
+        )
+        if handoff_hit:
+            raise ExternalReplyThreadingError(
+                "External reply threading references a HANDOFF message"
+            )
 
 
 def build_external_reply_plan(
@@ -50,9 +220,31 @@ def build_external_reply_plan(
     body_html: str | None = None,
     subject_override: str | None = None,
     recipients: list[str],
+    parent_internet_id_override: str | None = None,
 ) -> ExternalReplyPlan:
+    parent_internet_id = canonical_message_id(
+        parent_internet_id_override
+    ) or internet_message_id_of(parent_external_message)
+    if not parent_internet_id:
+        raise ExternalReplyThreadingError(
+            "External reply parent is missing internet_message_id"
+        )
+    scope = parent_external_message.get("message_scope")
+    if scope == MESSAGE_SCOPE_HANDOFF:
+        raise ExternalReplyThreadingError(
+            "External reply parent must not be a HANDOFF-scoped message"
+        )
     threading_fields = threading_fields_from_parent(parent_external_message)
-    parent_internet_id = internet_message_id_of(parent_external_message)
+    in_reply_to = parent_internet_id or threading_fields.get("in_reply_to")
+    references = threading_fields.get("references", [])
+    if parent_internet_id:
+        references = build_references_header(
+            [ref for ref in references if folded_message_id(ref) != folded_message_id(parent_internet_id)],
+            parent_internet_id,
+        ).split()
+    elif not references and in_reply_to:
+        references = [in_reply_to]
+
     if subject_override:
         subject = subject_override
     else:
@@ -67,8 +259,8 @@ def build_external_reply_plan(
         body_text=body_text,
         body_html=body_html,
         threading=ExternalReplyThreading(
-            in_reply_to=threading_fields.get("in_reply_to"),
-            references=threading_fields.get("references", []),
+            in_reply_to=in_reply_to,
+            references=references,
             parent_internal_message_id=parent_external_message["_id"],
             replying_to_internet_message_id=parent_internet_id,
         ),
@@ -76,6 +268,33 @@ def build_external_reply_plan(
     )
 
 
+def log_original_external_message(
+    *,
+    internal_id: ObjectId | str,
+    thread_id: ObjectId | str | None,
+    internet_message_id: str | None,
+    sender: str | None,
+    recipient: str | None,
+    subject: str | None,
+    in_reply_to: str | None,
+    references: list[str] | None,
+    raw_message_available: bool,
+) -> None:
+    logger.info(
+        "original_external_message internal_id=%s thread_id=%s internet_message_id=%s "
+        "from=%s to=%s subject=%s in_reply_to=%s references=%s raw_message_available=%s",
+        internal_id,
+        thread_id,
+        internet_message_id,
+        sender,
+        recipient,
+        subject,
+        in_reply_to,
+        references,
+        raw_message_available,
+    )
+
+
 def log_smtp_message_sending(
     *,
     generated_message_id: str | None,
@@ -114,7 +333,30 @@ async def send_external_reply(
     plan: ExternalReplyPlan,
     domain: MailDomain,
     resolution_method: str = "in_reply_to",
+    generated_by: str | None = None,
+    generated_from_handoff_message_id: ObjectId | None = None,
+    message_origin: str | None = None,
 ) -> dict:
+    if generated_from_handoff_message_id:
+        existing = await database.messages.find_one(
+            {
+                "generated_from_handoff_message_id": generated_from_handoff_message_id,
+                "message_scope": MESSAGE_SCOPE_EXTERNAL,
+            }
+        )
+        if existing:
+            logger.info(
+                "external_reply_idempotent_skip handoff_message=%s external=%s",
+                generated_from_handoff_message_id,
+                existing["_id"],
+            )
+            return existing
+
+    await assert_external_threading_chain(
+        database,
+        in_reply_to=plan.threading.in_reply_to,
+        references=plan.threading.references,
+    )
     log_smtp_message_sending(
         generated_message_id=None,
         in_reply_to=plan.threading.in_reply_to,
@@ -139,7 +381,8 @@ async def send_external_reply(
         in_reply_to=plan.threading.in_reply_to,
         references=plan.threading.references,
     )
-    normalized_id = normalize_message_id(internet_message_id)
+    normalized_id = canonical_message_id(internet_message_id)
+    id_fields = message_id_fields(normalized_id)
     log_smtp_message_sending(
         generated_message_id=normalized_id,
         in_reply_to=plan.threading.in_reply_to,
@@ -167,16 +410,23 @@ async def send_external_reply(
         "created_at": now,
         "sent_at": now,
         "status": "accepted",
-        "message_id": normalized_id,
-        "internet_message_id": normalized_id,
+        **id_fields,
         "in_reply_to": plan.threading.in_reply_to,
         "references": plan.threading.references,
         "parent_message_id": plan.threading.parent_internal_message_id,
         "thread_relation": "reply",
         "thread_resolution_method": resolution_method,
     }
+    if generated_by:
+        document["generated_by"] = generated_by
+    if generated_from_handoff_message_id:
+        document["generated_from_handoff_message_id"] = generated_from_handoff_message_id
+    if message_origin:
+        document["message_origin"] = message_origin
     result = await database.messages.insert_one(document)
     document["_id"] = result.inserted_id
+    from app.services.thread_service import attach_message_to_thread
+
     await attach_message_to_thread(
         database,
         account_id=account_id,

backend/app/services/handoff_mirror_policy.py

@@ -0,0 +1,62 @@
+"""Decide whether external messages should be mirrored into a Human Handoff thread."""
+
+MESSAGE_SCOPE_EXTERNAL = "EXTERNAL"
+
+# Message provenance — who authored / generated the content
+MESSAGE_ORIGIN_EXTERNAL_CUSTOMER = "EXTERNAL_CUSTOMER"
+MESSAGE_ORIGIN_HUMAN_OWNER = "HUMAN_OWNER"
+MESSAGE_ORIGIN_AGENT = "AGENT"
+MESSAGE_ORIGIN_SYSTEM = "SYSTEM"
+MESSAGE_ORIGIN_HANDOFF_MIRROR = "HANDOFF_MIRROR"
+
+GENERATED_BY_CUSTOMER = "CUSTOMER"
+GENERATED_BY_HUMAN = "HUMAN"
+GENERATED_BY_AGENT = "AGENT"
+GENERATED_BY_HANDOFF_BRIDGE = "HANDOFF_BRIDGE"
+GENERATED_BY_SYSTEM = "SYSTEM"
+GENERATED_BY_RELAY_ACK = "RELAY_ACK"
+
+RELAY_STATUS_SENT = "SENT"
+RELAY_STATUS_FAILED = "FAILED"
+RELAY_STATUS_PENDING = "PENDING"
+
+
+def should_mirror_to_human(message: dict) -> bool:
+    """Return True only for new external customer content worth mirroring."""
+    scope = message.get("message_scope", MESSAGE_SCOPE_EXTERNAL)
+    if scope != MESSAGE_SCOPE_EXTERNAL:
+        return False
+
+    if message.get("direction") != "inbound":
+        return False
+
+    generated_by = message.get("generated_by")
+    if generated_by in (
+        GENERATED_BY_HANDOFF_BRIDGE,
+        GENERATED_BY_RELAY_ACK,
+        GENERATED_BY_SYSTEM,
+    ):
+        return False
+
+    origin = message.get("message_origin")
+    if origin in (
+        MESSAGE_ORIGIN_HUMAN_OWNER,
+        MESSAGE_ORIGIN_SYSTEM,
+        MESSAGE_ORIGIN_HANDOFF_MIRROR,
+    ):
+        return False
+
+    if message.get("generated_from_handoff_message_id"):
+        return False
+
+    return True
+
+
+def is_relay_acknowledgment_message(message: dict) -> bool:
+    """Detect legacy or internal delivery-receipt messages in handoff threads."""
+    if message.get("generated_by") == GENERATED_BY_RELAY_ACK:
+        return True
+    if message.get("message_origin") == MESSAGE_ORIGIN_SYSTEM:
+        return True
+    body = (message.get("body_text") or message.get("snippet") or "").strip()
+    return message.get("direction") == "outbound" and body.startswith("Sent to customer")

backend/app/services/human_handoff_service.py

@@ -5,18 +5,37 @@ from datetime import datetime, timezone
 from bson import ObjectId
 
 from app.services.external_reply_service import (
+    ExternalReplyThreadingError,
     build_external_reply_plan,
+    internet_message_id_of,
+    log_original_external_message,
+    resolve_external_reply_parent,
+    resolve_internet_message_id,
     send_external_reply,
 )
 from app.services.handoff_email_templates import (
-    build_relay_confirmation_text,
     external_reply_subject,
     handoff_email_subject,
 )
+from app.services.handoff_mirror_policy import (
+    GENERATED_BY_HANDOFF_BRIDGE,
+    GENERATED_BY_HUMAN,
+    GENERATED_BY_SYSTEM,
+    MESSAGE_ORIGIN_HANDOFF_MIRROR,
+    MESSAGE_ORIGIN_HUMAN_OWNER,
+    RELAY_STATUS_FAILED,
+    RELAY_STATUS_SENT,
+    should_mirror_to_human,
+)
 from app.services.handoff_mirror import load_mirror_payload, load_mirror_update_payload
 from app.services.handoff_evaluator import evaluate_handoff
 from app.services.mail_sender import deliver_email
-from app.services.message_ids import normalize_message_id, parse_message_id_list
+from app.services.message_ids import (
+    canonical_message_id,
+    folded_message_id,
+    message_id_fields,
+    parse_message_id_list,
+)
 from app.services.reply_extractor import extract_new_reply_content, validate_external_reply_body
 from app.services.thread_service import (
     attach_message_to_thread,
@@ -226,9 +245,13 @@ async def resolve_handoff_from_inbound(
         return None
 
     for internet_id in reversed(parse_message_id_list(" ".join(search_ids))):
+        folded = folded_message_id(internet_id)
         message = await database.messages.find_one(
             {
-                "message_id": internet_id,
+                "$or": [
+                    {"message_id_folded": folded},
+                    {"message_id": internet_id},
+                ],
                 "message_scope": MESSAGE_SCOPE_HANDOFF,
             }
         )
@@ -256,6 +279,26 @@ async def resolve_handoff_from_inbound(
     return None
 
 
+async def _handoff_mirror_exists(database, external_message_id: ObjectId) -> bool:
+    existing = await database.messages.find_one(
+        {
+            "mirrored_from_external_message_id": external_message_id,
+            "message_scope": MESSAGE_SCOPE_HANDOFF,
+        },
+        {"_id": 1},
+    )
+    return existing is not None
+
+
+async def _bridged_external_exists(database, handoff_message_id: ObjectId) -> dict | None:
+    return await database.messages.find_one(
+        {
+            "generated_from_handoff_message_id": handoff_message_id,
+            "message_scope": MESSAGE_SCOPE_EXTERNAL,
+        }
+    )
+
+
 async def store_handoff_message(
     database,
     *,
@@ -277,6 +320,11 @@ async def store_handoff_message(
     raw_body: str | None = None,
     new_reply_content: str | None = None,
     body_html: str | None = None,
+    message_origin: str | None = None,
+    generated_by: str | None = None,
+    mirrored_from_external_message_id: ObjectId | None = None,
+    relay_status: str | None = None,
+    generated_external_message_id: ObjectId | None = None,
 ) -> dict:
     now = event_at or _now()
     display_body = new_reply_content if new_reply_content is not None else body_text
@@ -297,11 +345,20 @@ async def store_handoff_message(
         "sent_at": now if direction == "outbound" else None,
         "received_at": now if direction == "inbound" else None,
         "status": "received" if direction == "inbound" else "accepted",
-        "message_id": normalize_message_id(internet_message_id),
-        "internet_message_id": normalize_message_id(internet_message_id),
-        "in_reply_to": normalize_message_id(in_reply_to),
+        **message_id_fields(internet_message_id),
+        "in_reply_to": canonical_message_id(in_reply_to),
         "references": references or [],
     }
+    if message_origin:
+        document["message_origin"] = message_origin
+    if generated_by:
+        document["generated_by"] = generated_by
+    if mirrored_from_external_message_id:
+        document["mirrored_from_external_message_id"] = mirrored_from_external_message_id
+    if relay_status:
+        document["relay_status"] = relay_status
+    if generated_external_message_id:
+        document["generated_external_message_id"] = generated_external_message_id
     result = await database.messages.insert_one(document)
     document["_id"] = result.inserted_id
     await attach_message_to_thread(
@@ -357,6 +414,23 @@ async def send_handoff_to_human(
     domain,
     existing_handoff: dict | None = None,
 ) -> dict:
+    if await _handoff_mirror_exists(database, external_message["_id"]):
+        logger.info(
+            "handoff_mirror_skipped_duplicate external_message=%s",
+            external_message["_id"],
+        )
+        if existing_handoff:
+            return existing_handoff
+        existing = await database.human_handoffs.find_one(
+            {
+                "account_id": account_id,
+                "external_thread_id": external_thread_id,
+            }
+        )
+        if existing:
+            return existing
+        raise RuntimeError("Handoff mirror duplicate without handoff record")
+
     external_subject = external_message.get("subject", "")
     stable_subject = handoff_email_subject(external_subject)
     mirror_source = {**external_message, "_agent_email": agent_email}
@@ -394,6 +468,8 @@ async def send_handoff_to_human(
             "updated_at": _now(),
             "last_external_message_at": message_event_at(external_message),
             "last_human_message_at": None,
+            "trigger_external_message_id": external_message["_id"],
+            "trigger_internet_message_id": internet_message_id_of(external_message),
         }
         insert = await database.human_handoffs.insert_one(handoff_doc)
         handoff_id = insert.inserted_id
@@ -431,6 +507,9 @@ async def send_handoff_to_human(
         handoff_id=handoff_id,
         link_relationship=relationship,
         external_message_id=external_message["_id"],
+        message_origin=MESSAGE_ORIGIN_HANDOFF_MIRROR,
+        generated_by=GENERATED_BY_SYSTEM,
+        mirrored_from_external_message_id=external_message["_id"],
     )
 
     await database.human_handoffs.update_one(
@@ -504,7 +583,8 @@ async def process_human_reply(
         in_reply_to=inbound_parsed.get("in_reply_to"),
         references=inbound_parsed.get("references", []),
         handoff_id=handoff["_id"],
-        link_relationship=LINK_HUMAN_TRIGGERED_EXTERNAL,
+        message_origin=MESSAGE_ORIGIN_HUMAN_OWNER,
+        generated_by=GENERATED_BY_HUMAN,
     )
 
     now = _now()
@@ -550,22 +630,68 @@ async def process_human_reply(
         return
 
     external_recipient = handoff.get("external_sender") or ""
-    external_parent = await _latest_external_customer_message(
+    external_parent = await resolve_external_reply_parent(
         database,
-        account_id,
-        external_thread_id,
-        external_recipient,
+        account_id=account_id,
+        external_thread_id=external_thread_id,
+        customer_address=external_recipient,
+        handoff=handoff,
     )
     if external_parent is None:
-        external_parent = await _latest_external_message(database, account_id, external_thread_id)
-    if external_parent is None:
-        logger.warning("handoff_reply_missing_external_parent handoff_id=%s", handoff["_id"])
+        logger.warning(
+            "handoff_reply_missing_external_parent handoff_id=%s external_thread=%s "
+            "trigger_internet_message_id=%s",
+            handoff["_id"],
+            external_thread_id,
+            handoff.get("trigger_internet_message_id"),
+        )
         return
 
-    external_recipient = external_recipient or external_parent.get("sender", {}).get("address")
+    parent_sender = (external_parent.get("sender") or {}).get("address")
+    log_original_external_message(
+        internal_id=external_parent["_id"],
+        thread_id=external_thread_id,
+        internet_message_id=internet_message_id_of(external_parent),
+        sender=parent_sender,
+        recipient=agent_email,
+        subject=external_parent.get("subject"),
+        in_reply_to=external_parent.get("in_reply_to"),
+        references=external_parent.get("references"),
+        raw_message_available=bool(external_parent.get("raw_mime")),
+    )
+
+    external_recipient = external_recipient or parent_sender
     if not external_recipient:
         return
 
+    parent_internet_id = await resolve_internet_message_id(database, external_parent)
+    if not parent_internet_id:
+        logger.error(
+            "handoff_reply_missing_parent_message_id handoff_id=%s parent_internal_id=%s",
+            handoff["_id"],
+            external_parent["_id"],
+        )
+        return
+
+    logger.info(
+        "handoff_external_reply_threading handoff_id=%s parent_internal_id=%s "
+        "parent_internet_message_id=%s human_in_reply_to=%s human_references=%s",
+        handoff["_id"],
+        external_parent["_id"],
+        parent_internet_id,
+        human_message.get("in_reply_to"),
+        human_message.get("references"),
+    )
+
+    existing_bridge = await _bridged_external_exists(database, human_message["_id"])
+    if existing_bridge:
+        logger.info(
+            "handoff_bridge_already_processed handoff_message=%s external=%s",
+            human_message["_id"],
+            existing_bridge["_id"],
+        )
+        return
+
     customer_body = new_reply_content
     reply_plan = build_external_reply_plan(
         parent_external_message=external_parent,
@@ -574,7 +700,15 @@ async def process_human_reply(
             handoff.get("external_subject") or external_parent.get("subject", "")
         ),
         recipients=[external_recipient],
+        parent_internet_id_override=parent_internet_id,
     )
+    logger.info(
+        "handoff_external_reply_headers handoff_id=%s b_in_reply_to=%s b_references=%s",
+        handoff["_id"],
+        reply_plan.threading.in_reply_to,
+        reply_plan.threading.references,
+    )
+    try:
         external_doc = await send_external_reply(
             database,
             account_id=account_id,
@@ -584,52 +718,69 @@ async def process_human_reply(
             plan=reply_plan,
             domain=domain,
             resolution_method="human_handoff_bridge",
+            generated_by=GENERATED_BY_HANDOFF_BRIDGE,
+            generated_from_handoff_message_id=human_message["_id"],
         )
+    except ExternalReplyThreadingError:
+        await database.messages.update_one(
+            {"_id": human_message["_id"]},
+            {"$set": {"relay_status": RELAY_STATUS_FAILED, "relay_failed_at": now}},
+        )
+        logger.error(
+            "handoff_bridge_threading_rejected handoff_id=%s handoff_message=%s "
+            "parent_internet_message_id=%s",
+            handoff["_id"],
+            human_message["_id"],
+            parent_internet_id,
+        )
+        return
+    except Exception:
+        await database.messages.update_one(
+            {"_id": human_message["_id"]},
+            {"$set": {"relay_status": RELAY_STATUS_FAILED, "relay_failed_at": now}},
+        )
+        logger.exception(
+            "handoff_bridge_send_failed handoff_id=%s handoff_message=%s",
+            handoff["_id"],
+            human_message["_id"],
+        )
+        raise
 
-    relay_body = build_relay_confirmation_text(customer_body)
-    relay_subject = _handoff_reply_subject(handoff.get("external_subject", ""))
-    relay_threading = threading_fields_from_parent(human_message)
-    relay_internet_id = await deliver_email(
-        sender=agent_email,
-        recipients=[human_email],
-        cc=[],
-        bcc=[],
-        subject=relay_subject,
-        body_text=relay_body,
-        body_html=None,
-        domain=domain,
-        in_reply_to=relay_threading.get("in_reply_to"),
-        references=relay_threading.get("references"),
+    await database.messages.update_one(
+        {"_id": human_message["_id"]},
+        {
+            "$set": {
+                "relay_status": RELAY_STATUS_SENT,
+                "generated_external_message_id": external_doc["_id"],
+                "relay_delivered_at": now,
+            }
+        },
     )
-    await store_handoff_message(
+    await _record_message_link(
         database,
-        account_id=account_id,
-        agent_email=agent_email,
-        handoff_thread_id=handoff_thread_id,
-        direction="outbound",
-        sender={"name": "Gentic Agent", "address": agent_email},
-        recipients=[{"name": human_email.split("@")[0], "address": human_email}],
-        subject=relay_subject,
-        body_text=relay_body,
-        internet_message_id=relay_internet_id,
-        in_reply_to=relay_threading.get("in_reply_to"),
-        references=relay_threading.get("references"),
         handoff_id=handoff["_id"],
-        link_relationship=LINK_EXTERNAL_RELAYED,
-        external_message_id=ext_result.inserted_id,
+        external_message_id=external_doc["_id"],
+        handoff_message_id=human_message["_id"],
+        relationship=LINK_HUMAN_TRIGGERED_EXTERNAL,
     )
 
     await database.human_handoffs.update_one(
         {"_id": handoff["_id"]},
         {
             "$set": {
-                "status": HANDOFF_STATUS_AGENT_REPLIED,
-                "current_owner": OWNER_AGENT,
+                "status": HANDOFF_STATUS_WAITING,
+                "current_owner": OWNER_HUMAN,
                 "updated_at": now,
                 "last_human_message_at": now,
             }
         },
     )
+    logger.info(
+        "handoff_bridge_complete handoff_id=%s handoff_message=%s external=%s",
+        handoff["_id"],
+        human_message["_id"],
+        external_doc["_id"],
+    )
 
 
 async def maybe_trigger_handoff(
@@ -641,6 +792,40 @@ async def maybe_trigger_handoff(
     external_thread_id: ObjectId,
     domain,
 ) -> None:
+    if not should_mirror_to_human(external_message):
+        logger.debug(
+            "handoff_mirror_skipped message_id=%s generated_by=%s origin=%s direction=%s",
+            external_message.get("_id"),
+            external_message.get("generated_by"),
+            external_message.get("message_origin"),
+            external_message.get("direction"),
+        )
+        return
+
+    existing = await find_active_handoff_for_external_thread(
+        database, account_id, external_thread_id
+    )
+    if existing:
+        if await _handoff_mirror_exists(database, external_message["_id"]):
+            return
+        rule = await _handoff_rule_for_human(
+            database, account_id, existing.get("human_email", "")
+        )
+        if rule and not rule.get("allow_replies", True):
+            return
+        await send_handoff_to_human(
+            database,
+            account_id=account_id,
+            agent_email=agent_email,
+            human_email=existing["human_email"],
+            external_message=external_message,
+            external_thread_id=external_thread_id,
+            handoff_reason=existing.get("handoff_reason", ""),
+            domain=domain,
+            existing_handoff=existing,
+        )
+        return
+
     rules = await database.human_handoff_rules.find(
         {"account_id": account_id, "active": True}
     ).to_list(length=20)
@@ -667,9 +852,6 @@ async def maybe_trigger_handoff(
         if not rule.get("allow_replies", True):
             continue
 
-        existing = await find_active_handoff_for_external_thread(
-            database, account_id, external_thread_id
-        )
         reason = (rule.get("instructions") or "").strip() or "This conversation requires human attention."
         await send_handoff_to_human(
             database,
@@ -680,6 +862,6 @@ async def maybe_trigger_handoff(
             external_thread_id=external_thread_id,
             handoff_reason=reason,
             domain=domain,
-            existing_handoff=existing,
+            existing_handoff=None,
         )
         return

backend/app/services/inbound_smtp.py

@@ -10,6 +10,10 @@ from app.config import get_settings
 from app.database import SessionFactory, mongo_db
 from app.models import Account
 from app.services.domain_provisioning import get_domain_by_name
+from app.services.handoff_mirror_policy import (
+    GENERATED_BY_CUSTOMER,
+    MESSAGE_ORIGIN_EXTERNAL_CUSTOMER,
+)
 from app.services.human_handoff_service import (
     MESSAGE_SCOPE_EXTERNAL,
     MESSAGE_SCOPE_HANDOFF,
@@ -17,7 +21,8 @@ from app.services.human_handoff_service import (
     resolve_handoff_from_inbound,
     maybe_trigger_handoff,
 )
-from app.services.message_ids import normalize_message_id
+from app.services.external_reply_service import log_original_external_message
+from app.services.message_ids import canonical_message_id, message_id_fields
 from app.services.mime_parser import parse_mime
 from app.services.raw_mime_storage import delivery_digest, get_raw_mime_storage
 from app.services.thread_resolver import participants_from_parsed, resolve_thread
@@ -128,7 +133,8 @@ class InboundHandler:
             stored = await get_raw_mime_storage().put(raw, delivery_key)
             now = datetime.now(timezone.utc)
             sender_address = (parsed.get("sender") or {}).get("address", "").lower()
-            internet_message_id = normalize_message_id(parsed.get("message_id"))
+            internet_message_id = canonical_message_id(parsed.get("message_id"))
+            id_fields = message_id_fields(internet_message_id)
 
             for recipient in recipients:
                 account_id = envelope.account_ids[recipient]
@@ -160,8 +166,7 @@ class InboundHandler:
                         "delivery_key": delivery_key,
                         "received_at": now,
                         "created_at": now,
-                        "message_id": internet_message_id,
-                        "internet_message_id": internet_message_id,
+                        **id_fields,
                     }
                     await process_human_reply(
                         mongo_db,
@@ -194,8 +199,7 @@ class InboundHandler:
                 "envelope_sender": envelope.mail_from.lower(),
                 "envelope_recipients": recipients,
                 "delivery_key": delivery_key,
-                "message_id": internet_message_id,
-                "internet_message_id": internet_message_id,
+                **id_fields,
                 "raw_mime": {
                     "bucket": stored.bucket,
                     "name": stored.name,
@@ -252,14 +256,15 @@ class InboundHandler:
                     {
                         "$set": {
                             "message_scope": MESSAGE_SCOPE_EXTERNAL,
-                            "message_id": internet_message_id,
-                            "internet_message_id": internet_message_id,
-                            "in_reply_to": normalize_message_id(parsed.get("in_reply_to")),
+                            **id_fields,
+                            "in_reply_to": canonical_message_id(parsed.get("in_reply_to")),
                             "references": parsed.get("references", []),
                             "raw_headers": parsed.get("raw_headers", {}),
                             "parent_message_id": resolution.parent_message_id,
                             "thread_relation": resolution.thread_relation,
                             "thread_resolution_method": resolution.thread_resolution_method,
+                            "message_origin": MESSAGE_ORIGIN_EXTERNAL_CUSTOMER,
+                            "generated_by": GENERATED_BY_CUSTOMER,
                         }
                     },
                 )
@@ -298,6 +303,14 @@ class InboundHandler:
                         database_session,
                         agent_email.split("@", 1)[1],
                     )
+                persisted.update(
+                    {
+                        "message_scope": MESSAGE_SCOPE_EXTERNAL,
+                        "direction": "inbound",
+                        "message_origin": MESSAGE_ORIGIN_EXTERNAL_CUSTOMER,
+                        "generated_by": GENERATED_BY_CUSTOMER,
+                    }
+                )
                 if domain is not None:
                     await maybe_trigger_handoff(
                         mongo_db,
@@ -307,6 +320,17 @@ class InboundHandler:
                         external_thread_id=thread_id,
                         domain=domain,
                     )
+                log_original_external_message(
+                    internal_id=persisted["_id"],
+                    thread_id=thread_id,
+                    internet_message_id=internet_message_id,
+                    sender=sender_address,
+                    recipient=recipient,
+                    subject=parsed.get("subject"),
+                    in_reply_to=canonical_message_id(parsed.get("in_reply_to")),
+                    references=parsed.get("references", []),
+                    raw_message_available=bool(persisted.get("raw_mime")),
+                )
                 log_smtp_message_received(
                     message_id=internet_message_id,
                     in_reply_to=parsed.get("in_reply_to"),

backend/app/services/mail_sender.py

@@ -1,5 +1,7 @@
 import email.policy
 import email.utils
+import logging
+from email.parser import BytesParser
 from email.message import EmailMessage, Message
 from email.mime.base import MIMEBase
 from email.mime.multipart import MIMEMultipart
@@ -13,6 +15,20 @@ from app.models import MailDomain
 from app.services.mime_body_extractor import ExtractedMimePart
 
 SMTP_POLICY = email.policy.SMTP.clone(max_line_length=998)
+logger = logging.getLogger(__name__)
+
+
+def log_outbound_mime_headers(message: Message) -> None:
+    logger.info(
+        "smtp_outbound_mime_headers message_id=%s in_reply_to=%s references=%s "
+        "from=%s to=%s subject=%s",
+        message.get("Message-ID"),
+        message.get("In-Reply-To"),
+        message.get("References"),
+        message.get("From"),
+        message.get("To"),
+        message.get("Subject"),
+    )
 
 
 def _mime_attachment(part: ExtractedMimePart) -> MIMEBase:
@@ -60,7 +76,9 @@ def build_message(
         root["Message-ID"] = generated_id
         if in_reply_to:
             root["In-Reply-To"] = in_reply_to
-        if references:
+            refs = references or [in_reply_to]
+            root["References"] = " ".join(refs)
+        elif references:
             root["References"] = " ".join(references)
 
         alternative = MIMEMultipart("alternative", policy=SMTP_POLICY)
@@ -94,7 +112,9 @@ def build_message(
     message["Message-ID"] = generated_id
     if in_reply_to:
         message["In-Reply-To"] = in_reply_to
-    if references:
+        refs = references or [in_reply_to]
+        message["References"] = " ".join(refs)
+    elif references:
         message["References"] = " ".join(references)
     message.set_content(body_text)
     if body_html:
@@ -115,6 +135,10 @@ def sign_message(message: Message, domain: MailDomain) -> bytes:
     ]
     if message.get("Cc"):
         signed_headers.insert(3, b"cc")
+    if message.get("In-Reply-To"):
+        signed_headers.append(b"in-reply-to")
+    if message.get("References"):
+        signed_headers.append(b"references")
     signature = dkim.sign(
         raw_message,
         domain.dkim_selector.encode(),
@@ -180,7 +204,9 @@ async def deliver_email(
         inline_parts=inline_parts,
         file_attachments=file_attachments,
     )
+    log_outbound_mime_headers(message)
     signed = sign_message(message, domain)
+    log_signed_mime_headers(signed)
     await send_via_smtp(
         signed,
         sender=sender,

backend/app/services/message_ids.py

@@ -3,7 +3,8 @@ import re
 _MESSAGE_ID_PATTERN = re.compile(r"<[^>]+>")
 
 
-def normalize_message_id(value: str | None) -> str | None:
+def canonical_message_id(value: str | None) -> str | None:
+    """Normalize to angle-bracket form while preserving original case."""
     if value is None:
         return None
     cleaned = value.strip()
@@ -11,22 +12,59 @@ def normalize_message_id(value: str | None) -> str | None:
         return None
     if not cleaned.startswith("<"):
         cleaned = f"<{cleaned.strip('<>')}>"
-    return cleaned.lower()
+    return cleaned
+
+
+def folded_message_id(value: str | None) -> str | None:
+    """Case-insensitive lookup key for Message-ID matching."""
+    canonical = canonical_message_id(value)
+    return canonical.casefold() if canonical else None
+
+
+def normalize_message_id(value: str | None) -> str | None:
+    """Preserve case — Gmail threading requires exact Message-ID tokens."""
+    return canonical_message_id(value)
+
+
+def internet_message_id_of_message(message: dict) -> str | None:
+    """Resolve Message-ID, preferring raw MIME header case when stored value was folded."""
+    stored = canonical_message_id(
+        message.get("internet_message_id") or message.get("message_id")
+    )
+    raw_header = (message.get("raw_headers") or {}).get("message_id")
+    raw_canonical = canonical_message_id(raw_header)
+    if raw_canonical and stored:
+        if folded_message_id(raw_canonical) == folded_message_id(stored):
+            return raw_canonical
+    return stored or raw_canonical
+
+
+def message_id_fields(value: str | None) -> dict[str, str]:
+    canonical = canonical_message_id(value)
+    if not canonical:
+        return {}
+    folded = folded_message_id(canonical)
+    return {
+        "message_id": canonical,
+        "internet_message_id": canonical,
+        "message_id_folded": folded or canonical.casefold(),
+    }
 
 
 def parse_message_id_list(header_value: str | None) -> list[str]:
     if not header_value:
         return []
-    folded = re.sub(r"\s+", " ", header_value.strip())
+    folded_header = re.sub(r"\s+", " ", header_value.strip())
     seen: set[str] = set()
     ordered: list[str] = []
-    for match in _MESSAGE_ID_PATTERN.findall(folded):
-        normalized = normalize_message_id(match)
-        if normalized and normalized not in seen:
-            seen.add(normalized)
-            ordered.append(normalized)
+    for match in _MESSAGE_ID_PATTERN.findall(folded_header):
+        canonical = canonical_message_id(match)
+        key = folded_message_id(canonical)
+        if canonical and key and key not in seen:
+            seen.add(key)
+            ordered.append(canonical)
     if not ordered:
-        single = normalize_message_id(folded)
+        single = canonical_message_id(folded_header)
         if single:
             ordered.append(single)
     return ordered
@@ -41,11 +79,13 @@ def build_references_header(references: list[str], parent_message_id: str) -> st
     chain: list[str] = []
     seen: set[str] = set()
     for value in references:
-        normalized = normalize_message_id(value)
-        if normalized and normalized not in seen:
-            seen.add(normalized)
-            chain.append(normalized)
-    parent = normalize_message_id(parent_message_id)
-    if parent and parent not in seen:
+        canonical = canonical_message_id(value)
+        key = folded_message_id(canonical)
+        if canonical and key and key not in seen:
+            seen.add(key)
+            chain.append(canonical)
+    parent = canonical_message_id(parent_message_id)
+    parent_key = folded_message_id(parent)
+    if parent and parent_key and parent_key not in seen:
         chain.append(parent)
     return " ".join(chain)

backend/app/services/mime_parser.py

@@ -28,6 +28,13 @@ def _text(part) -> str:
     return value if isinstance(value, str) else ""
 
 
+def _header_str(value) -> str | None:
+    if value is None:
+        return None
+    cleaned = str(value).strip()
+    return cleaned or None
+
+
 def _snippet(text: str, html: str | None, limit: int = 240) -> str:
     source = text
     if not source and html:
@@ -81,15 +88,15 @@ def parse_mime(raw: bytes) -> dict:
     in_reply_to = first_message_id(message.get("In-Reply-To"))
     references = parse_message_id_list(message.get("References"))
     raw_headers = {
-        "message_id": message.get("Message-ID"),
-        "in_reply_to": message.get("In-Reply-To"),
-        "references": message.get("References"),
-        "subject": message.get("Subject"),
-        "date": message.get("Date"),
-        "from": message.get("From"),
-        "to": message.get("To"),
-        "cc": message.get("Cc"),
-        "reply_to": message.get("Reply-To"),
+        "message_id": _header_str(message.get("Message-ID")),
+        "in_reply_to": _header_str(message.get("In-Reply-To")),
+        "references": _header_str(message.get("References")),
+        "subject": _header_str(message.get("Subject")),
+        "date": _header_str(message.get("Date")),
+        "from": _header_str(message.get("From")),
+        "to": _header_str(message.get("To")),
+        "cc": _header_str(message.get("Cc")),
+        "reply_to": _header_str(message.get("Reply-To")),
     }
     return {
         "message_id": internet_message_id,

backend/app/services/thread_resolver.py

@@ -5,8 +5,8 @@ from datetime import datetime, timedelta, timezone
 from bson import ObjectId
 
 from app.services.message_ids import (
-    first_message_id,
-    normalize_message_id,
+    canonical_message_id,
+    folded_message_id,
     parse_message_id_list,
 )
 from app.services.subject_normalize import normalize_subject
@@ -52,14 +52,17 @@ def resolve_thread_from_known(
     known_threads: dict[ObjectId, list[KnownMessage]] | None = None,
     now: datetime | None = None,
 ) -> ThreadResolution:
-    scoped = {
-        key: value
-        for key, value in known_by_internet_id.items()
-        if value.account_id == account_id
-    }
-    in_reply = normalize_message_id(in_reply_to)
-    if in_reply and in_reply in scoped:
-        parent = scoped[in_reply]
+    scoped: dict[str, KnownMessage] = {}
+    for key, value in known_by_internet_id.items():
+        if value.account_id != account_id:
+            continue
+        fold_key = folded_message_id(key) or folded_message_id(value.internet_message_id)
+        if fold_key:
+            scoped[fold_key] = value
+    in_reply = canonical_message_id(in_reply_to)
+    in_reply_folded = folded_message_id(in_reply)
+    if in_reply_folded and in_reply_folded in scoped:
+        parent = scoped[in_reply_folded]
         return ThreadResolution(
             thread_id=parent.thread_id,
             parent_message_id=parent.id,
@@ -68,9 +71,9 @@ def resolve_thread_from_known(
         )
 
     reference_ids = [
-        normalize_message_id(value)
+        folded_message_id(value)
         for value in references
-        if normalize_message_id(value)
+        if folded_message_id(value)
     ]
     for reference in reversed(reference_ids):
         if reference in scoped:
@@ -155,22 +158,31 @@ async def lookup_known_messages(
     account_id: str,
     internet_ids: list[str],
 ) -> dict[str, KnownMessage]:
-    normalized_ids = [
+    folded_ids = [
         value
-        for value in {normalize_message_id(item) for item in internet_ids}
+        for value in {folded_message_id(item) for item in internet_ids}
         if value
     ]
-    if not normalized_ids:
+    if not folded_ids:
         return {}
 
     pipeline = [
         {
             "$match": {
-                "message_id": {"$in": normalized_ids},
+                "$and": [
+                    {
+                        "$or": [
+                            {"message_id_folded": {"$in": folded_ids}},
+                            {"message_id": {"$in": folded_ids}},
+                        ]
+                    },
+                    {
                         "$or": [
                             {"message_scope": "EXTERNAL"},
                             {"message_scope": {"$exists": False}},
-                ],
+                        ]
+                    },
+                ]
             }
         },
         {
@@ -198,15 +210,18 @@ async def lookup_known_messages(
     known: dict[str, KnownMessage] = {}
     cursor = await database.messages.aggregate(pipeline)
     async for document in cursor:
-        internet_id = normalize_message_id(document.get("message_id"))
-        if not internet_id:
+        internet_id = canonical_message_id(
+            document.get("internet_message_id") or document.get("message_id")
+        )
+        internet_key = folded_message_id(internet_id)
+        if not internet_id or not internet_key:
             continue
         entry = document["entry"][0]
         thread_id = entry.get("thread_id") or document.get("thread_id")
         if thread_id is None:
             continue
         participants = participants_from_parsed(document)
-        known[internet_id] = KnownMessage(
+        known[internet_key] = KnownMessage(
             id=document["_id"],
             thread_id=thread_id,
             internet_message_id=internet_id,

backend/app/services/thread_service.py

@@ -2,7 +2,11 @@ from datetime import datetime, timezone
 
 from bson import ObjectId
 
-from app.services.message_ids import build_references_header, normalize_message_id
+from app.services.message_ids import (
+    build_references_header,
+    canonical_message_id,
+    internet_message_id_of_message,
+)
 from app.services.subject_normalize import normalize_subject
 from app.services.thread_resolver import ThreadResolution
 
@@ -121,9 +125,7 @@ async def ensure_thread_for_message(
 
 
 def threading_fields_from_parent(parent: dict) -> dict:
-    parent_internet_id = normalize_message_id(
-        parent.get("internet_message_id") or parent.get("message_id")
-    )
+    parent_internet_id = internet_message_id_of_message(parent)
     parent_references = parent.get("references") or []
     if isinstance(parent_references, str):
         parent_references = [parent_references]

backend/tests/test_external_reply_service.py

@@ -1,6 +1,10 @@
 from bson import ObjectId
 
-from app.services.external_reply_service import build_external_reply_plan
+from app.services.external_reply_service import (
+    build_external_reply_plan,
+    internet_message_id_of,
+)
+from app.services.message_ids import canonical_message_id, folded_message_id
 from app.services.thread_service import threading_fields_from_parent
 
 
@@ -55,3 +59,76 @@ def test_external_reply_never_uses_internal_object_id_as_in_reply_to():
     )
     assert str(parent["_id"]) not in (plan.threading.in_reply_to or "")
     assert all(str(parent["_id"]) not in ref for ref in plan.threading.references)
+
+
+def test_handoff_bridge_reply_uses_external_parent_not_human_reply():
+    """Human handoff must thread B against customer A, never private H2."""
+    customer_a = {
+        "_id": ObjectId(),
+        "internet_message_id": "<CAHabc123@mail.gmail.com>",
+        "message_id": "<CAHabc123@mail.gmail.com>",
+        "message_scope": "EXTERNAL",
+        "direction": "inbound",
+        "references": [],
+        "subject": "fresh test",
+    }
+    human_h2 = {
+        "_id": ObjectId(),
+        "internet_message_id": "<H2@gmail.com>",
+        "message_id": "<H2@gmail.com>",
+        "message_scope": "HANDOFF",
+        "direction": "inbound",
+        "in_reply_to": "<H1@gentic.email>",
+        "references": ["<H1@gentic.email>"],
+        "subject": "Re: fresh test",
+    }
+    plan = build_external_reply_plan(
+        parent_external_message=customer_a,
+        body_text="human said B",
+        recipients=["091rabbytes@gmail.com"],
+    )
+    assert plan.threading.in_reply_to == "<CAHabc123@mail.gmail.com>"
+    assert plan.threading.references == ["<CAHabc123@mail.gmail.com>"]
+    assert human_h2["internet_message_id"] not in (plan.threading.in_reply_to or "")
+    assert all(
+        human_h2["internet_message_id"] not in ref for ref in plan.threading.references
+    )
+    assert "<H1@gentic.email>" not in plan.threading.references
+
+
+def test_raw_header_restores_gmail_message_id_case():
+    parent = {
+        "_id": ObjectId(),
+        "internet_message_id": "<cahabc123@mail.gmail.com>",
+        "message_id": "<cahabc123@mail.gmail.com>",
+        "raw_headers": {"message_id": "<CAHabc123@mail.gmail.com>"},
+        "references": [],
+        "subject": "fresh test",
+    }
+    plan = build_external_reply_plan(
+        parent_external_message=parent,
+        body_text="human said B",
+        recipients=["customer@gmail.com"],
+    )
+    assert plan.threading.in_reply_to == "<CAHabc123@mail.gmail.com>"
+    assert plan.threading.references == ["<CAHabc123@mail.gmail.com>"]
+
+
+def test_gmail_message_id_case_is_preserved_for_threading():
+    raw = "<CAHabc123@mail.gmail.com>"
+    assert canonical_message_id(raw) == raw
+    assert folded_message_id(raw) == "<cahabc123@mail.gmail.com>"
+    parent = {
+        "_id": ObjectId(),
+        "internet_message_id": raw,
+        "message_id": raw,
+        "references": [],
+        "subject": "fresh test",
+    }
+    plan = build_external_reply_plan(
+        parent_external_message=parent,
+        body_text="human said B",
+        recipients=["customer@gmail.com"],
+    )
+    assert plan.threading.in_reply_to == raw
+    assert internet_message_id_of(parent) == raw

backend/tests/test_handoff_mirror_policy.py

@@ -0,0 +1,57 @@
+from bson import ObjectId
+
+from app.services.handoff_mirror_policy import (
+    GENERATED_BY_CUSTOMER,
+    GENERATED_BY_HANDOFF_BRIDGE,
+    GENERATED_BY_RELAY_ACK,
+    MESSAGE_ORIGIN_EXTERNAL_CUSTOMER,
+    MESSAGE_ORIGIN_HUMAN_OWNER,
+    is_relay_acknowledgment_message,
+    should_mirror_to_human,
+)
+
+
+def test_should_mirror_customer_inbound():
+    message = {
+        "message_scope": "EXTERNAL",
+        "direction": "inbound",
+        "generated_by": GENERATED_BY_CUSTOMER,
+        "message_origin": MESSAGE_ORIGIN_EXTERNAL_CUSTOMER,
+    }
+    assert should_mirror_to_human(message) is True
+
+
+def test_should_not_mirror_handoff_bridge_outbound():
+    message = {
+        "message_scope": "EXTERNAL",
+        "direction": "outbound",
+        "generated_by": GENERATED_BY_HANDOFF_BRIDGE,
+        "generated_from_handoff_message_id": ObjectId(),
+    }
+    assert should_mirror_to_human(message) is False
+
+
+def test_should_not_mirror_human_origin():
+    message = {
+        "message_scope": "EXTERNAL",
+        "direction": "inbound",
+        "message_origin": MESSAGE_ORIGIN_HUMAN_OWNER,
+    }
+    assert should_mirror_to_human(message) is False
+
+
+def test_is_relay_acknowledgment_detects_legacy_body():
+    message = {
+        "direction": "outbound",
+        "body_text": "Sent to customer:\n\nhuman said B",
+    }
+    assert is_relay_acknowledgment_message(message) is True
+
+
+def test_is_relay_acknowledgment_detects_generated_by():
+    message = {
+        "direction": "outbound",
+        "generated_by": GENERATED_BY_RELAY_ACK,
+        "body_text": "anything",
+    }
+    assert is_relay_acknowledgment_message(message) is True

backend/tests/test_handoff_threading.py

@@ -0,0 +1,82 @@
+"""Integration-style tests for Human Handoff → Customer RFC threading."""
+
+from bson import ObjectId
+
+from app.services.external_reply_service import (
+    build_external_reply_plan,
+    internet_message_id_of,
+)
+from app.services.mail_sender import build_message, sign_message
+from types import SimpleNamespace
+
+
+def test_handoff_bridge_first_reply_threads_against_customer_a():
+    customer_a_id = "<CAHabc123@mail.gmail.com>"
+    customer_a = {
+        "_id": ObjectId(),
+        "internet_message_id": customer_a_id,
+        "message_id": customer_a_id,
+        "message_scope": "EXTERNAL",
+        "direction": "inbound",
+        "references": [],
+        "subject": "fresh test",
+    }
+    plan = build_external_reply_plan(
+        parent_external_message=customer_a,
+        body_text="human said B",
+        recipients=["091rabbytes@gmail.com"],
+    )
+    assert plan.threading.in_reply_to == customer_a_id
+    assert plan.threading.references == [customer_a_id]
+    assert plan.threading.in_reply_to != "<H2@gmail.com>"
+    assert "<H1@gentic.email>" not in plan.threading.references
+
+
+def test_lowercased_stored_id_recovered_via_raw_headers():
+    customer_a_id = "<CAHabc123@mail.gmail.com>"
+    parent = {
+        "_id": ObjectId(),
+        "internet_message_id": "<cahabc123@mail.gmail.com>",
+        "message_id": "<cahabc123@mail.gmail.com>",
+        "raw_headers": {"message_id": customer_a_id},
+        "references": [],
+        "subject": "fresh test",
+    }
+    plan = build_external_reply_plan(
+        parent_external_message=parent,
+        body_text="human said B",
+        recipients=["customer@gmail.com"],
+        parent_internet_id_override=internet_message_id_of(parent),
+    )
+    assert plan.threading.in_reply_to == customer_a_id
+    assert plan.threading.references == [customer_a_id]
+
+
+def test_signed_mime_preserves_threading_headers():
+    customer_a_id = "<CAHabc123@mail.gmail.com>"
+    message = build_message(
+        sender="suyash@gentic.email",
+        recipients=["091rabbytes@gmail.com"],
+        cc=[],
+        subject="Re: fresh test",
+        body_text="human said B",
+        body_html=None,
+        in_reply_to=customer_a_id,
+        references=[customer_a_id],
+        message_id="<B@gentic.email>",
+    )
+    domain = SimpleNamespace(
+        domain_name="gentic.email",
+        dkim_selector="agmail",
+        dkim_private_key="private-key",
+    )
+    from unittest.mock import patch
+
+    with patch(
+        "app.services.mail_sender.dkim.sign",
+        return_value=b"DKIM-Signature: test\r\n",
+    ):
+        signed = sign_message(message, domain)
+
+    assert f"In-Reply-To: {customer_a_id}".encode() in signed
+    assert f"References: {customer_a_id}".encode() in signed

backend/tests/test_human_handoff_service.py

@@ -73,6 +73,7 @@ async def test_resolve_handoff_requires_owner_and_thread_context():
         {
             "_id": message_id,
             "message_id": "<handoff-1@gentic.email>",
+            "message_id_folded": "<handoff-1@gentic.email>",
             "message_scope": MESSAGE_SCOPE_HANDOFF,
         }
     ]

backend/tests/test_mail_sender.py

@@ -41,3 +41,50 @@ async def test_send_via_smtp_uses_configured_client_hostname():
         )
 
     assert send_mock.await_args.kwargs["local_hostname"] == "backend.gentic.email"
+
+
+def test_build_message_sets_in_reply_to_and_references():
+    message = build_message(
+        sender="suyash@gentic.email",
+        recipients=["customer@gmail.com"],
+        cc=[],
+        subject="Re: fresh test",
+        body_text="human said B",
+        body_html=None,
+        in_reply_to="<CAHabc123@mail.gmail.com>",
+        references=["<CAHabc123@mail.gmail.com>"],
+        message_id="<B@gentic.email>",
+    )
+    assert message["Message-ID"] == "<B@gentic.email>"
+    assert message["In-Reply-To"] == "<CAHabc123@mail.gmail.com>"
+    assert message["References"] == "<CAHabc123@mail.gmail.com>"
+    raw = message.as_bytes()
+    assert b"In-Reply-To: <CAHabc123@mail.gmail.com>" in raw
+    assert b"References: <CAHabc123@mail.gmail.com>" in raw
+
+
+def test_signed_mime_preserves_in_reply_to():
+    from unittest.mock import patch
+
+    message = build_message(
+        sender="suyash@gentic.email",
+        recipients=["customer@gmail.com"],
+        cc=[],
+        subject="Re: fresh test",
+        body_text="human said B",
+        body_html=None,
+        in_reply_to="<CAHabc123@mail.gmail.com>",
+        references=["<CAHabc123@mail.gmail.com>"],
+        message_id="<B@gentic.email>",
+    )
+    domain = SimpleNamespace(
+        domain_name="gentic.email",
+        dkim_selector="agmail",
+        dkim_private_key="private-key",
+    )
+    with patch(
+        "app.services.mail_sender.dkim.sign",
+        return_value=b"DKIM-Signature: test\r\n",
+    ):
+        signed = sign_message(message, domain)
+    assert b"In-Reply-To: <CAHabc123@mail.gmail.com>" in signed

frontend/src/components/HandoffConversation.jsx

@@ -83,10 +83,26 @@ export default function HandoffConversation({ handoffId, onBack }) {
         {conversation.messages.map((message) => (
           <article key={message.id} className={`handoff-message handoff-message-${message.role}`}>
             <header>
-              <strong>{message.role === "human" ? "HUMAN OWNER" : "AGENT"}</strong>
+              <strong>
+                {message.role === "human"
+                  ? "HUMAN OWNER"
+                  : message.role === "customer"
+                    ? "CUSTOMER"
+                    : "AGENT"}
+              </strong>
               <time>{formatTime(message.event_at)}</time>
             </header>
             <HandoffMessageBody message={message} />
+            {message.role === "human" && message.relay_status === "SENT" && (
+              <p className="handoff-relay-status handoff-relay-status-sent">
+                ✓ Sent to customer
+              </p>
+            )}
+            {message.role === "human" && message.relay_status === "FAILED" && (
+              <p className="handoff-relay-status handoff-relay-status-failed">
+                ⚠ Could not send to customer
+              </p>
+            )}
           </article>
         ))}
       </div>

frontend/src/styles.css

@@ -559,6 +559,10 @@ h1 em { color:var(--coral); font-style:normal; font-weight:500; }
 .handoff-message-human header strong { color:#11966a; }
 .handoff-message header time { color:#9a96a5; font-size:11px; }
 .handoff-message-body { color:#292536; font-size:13px; line-height:1.6; white-space:pre-wrap; }
+.handoff-message-customer header strong { color:#2563eb; }
+.handoff-relay-status { margin:10px 0 0; font-size:12px; }
+.handoff-relay-status-sent { color:#11966a; }
+.handoff-relay-status-failed { color:#c2410c; }
 .handoff-show-quoted { margin-top:8px; padding:0; border:0; background:transparent; color:#6d5bd0; font-size:11px; font-weight:600; cursor:pointer; }
 .handoff-quoted-body { margin:8px 0 0; padding:10px 12px; border:1px solid #ece9f2; border-radius:8px; background:#faf9fc; color:#77758a; font-size:11px; line-height:1.45; white-space:pre-wrap; overflow:auto; }
 .handoff-settings-intro { margin-bottom:18px; padding:18px 20px; border:1px solid #ece7ff; border-radius:14px; background:#fbf9ff; }