feat(injects,conversations): configurable inject header, urgent inject marks the turn it cut

src/beaver_gateway/core/conversations.py

@@ -39,6 +39,7 @@ from claude_agent_sdk import (
 )
 from sqlmodel import col, select
 
+from beaver_gateway.core import injects
 from beaver_gateway.core.conversation_store import load_messages
 from beaver_gateway.core.distill import (
     Digest,
@@ -52,7 +53,7 @@ from beaver_gateway.core.distill import (
     trim_summary,
     written_paths,
 )
-from beaver_gateway.core.injects import InjectQueue, inject_header
+from beaver_gateway.core.injects import InjectQueue
 from beaver_gateway.core.kinds import KINDS, Kind, as_kind
 from beaver_gateway.core.transcript import (
     messages_from_entries,
@@ -157,6 +158,8 @@ class ConversationTexts:
     """
 
     merge_prompt: str = _DEFAULT_MERGE_PROMPT
+    inject_header: Callable[[injects.InjectContext], str] = injects.inject_header
+    """Framing line(s) above each inject: who it is from, whether it cut a turn."""
     interrupted: str = "прервано"
     answered: str = "Пользователь ответил: {answer}"
     unanswered: str = (
@@ -857,6 +860,7 @@ class Conversations:
                 _log.info(
                     "conversation %s: interrupted for urgent inject", conv.external_id
                 )
+                await self._queue.mark_interrupting(item)
         self._ensure_worker(cast("int", conv.id))
         return item
 
@@ -1650,7 +1654,10 @@ class Conversations:
                 prompt += "\n\n" + _bundle(batch[1:])
         else:
             origin = "inject"
-            prompt = "\n\n".join(f"{inject_header(i.origin)}\n{i.text}" for i in batch)
+            prompt = "\n\n".join(
+                f"{self._texts.inject_header(injects.context_of(i))}\n{i.text}"
+                for i in batch
+            )
         seed = await self._pending_seed(conv)
         if seed:
             prompt = f"{seed}\n\n{prompt}"

src/beaver_gateway/core/injects.py

@@ -9,6 +9,7 @@ runs one worker per conversation over them. A row that is still
 
 from __future__ import annotations
 
+from dataclasses import dataclass
 from datetime import UTC, datetime
 from typing import TYPE_CHECKING, Literal, cast
 
@@ -23,14 +24,40 @@ if TYPE_CHECKING:
     from beaver_gateway.storage.db import Database
     from beaver_gateway.storage.models import Conversation
 
-__all__ = ["PRIORITY_RANK", "InjectQueue", "Priority", "inject_header"]
+__all__ = ["PRIORITY_RANK", "InjectContext", "InjectQueue", "Priority", "inject_header"]
 
 Priority = Literal["urgent", "user", "normal"]
 PRIORITY_RANK: dict[str, int] = {"urgent": 0, "user": 1, "normal": 2}
 
+INTERRUPTED_TURN = (
+    "[этот инжект прервал предыдущий тёрн: «Request interrupted» выше - "
+    "прерывание, не отказ от тулзы]"
+)
 
-def inject_header(origin: str) -> str:
+
-    return f"[инжект: {origin} - это не Бобёр, отвечать не нужно, голос не обязателен]"
+@dataclass(frozen=True, slots=True)
+class InjectContext:
+    """What a header renderer knows about one queued inject."""
+
+    origin: str
+    priority: str
+    interrupted_turn: bool
+
+
+def inject_header(ctx: InjectContext) -> str:
+    """Default framing; a setup overrides it via ``ConversationTexts.inject_header``."""
+    head = (
+        f"[инжект: {ctx.origin} - это не Бобёр, отвечать не нужно, голос не обязателен]"
+    )
+    return f"{head}\n{INTERRUPTED_TURN}" if ctx.interrupted_turn else head
+
+
+def context_of(item: InjectQueueItem) -> InjectContext:
+    return InjectContext(
+        origin=item.origin,
+        priority=item.priority,
+        interrupted_turn=bool(item.interrupted_turn),
+    )
 
 
 class InjectQueue:
@@ -83,6 +110,15 @@ class InjectQueue:
     ) -> None:
         await self._mark(items, status=status)
 
+    async def mark_interrupting(self, item: InjectQueueItem) -> None:
+        item.interrupted_turn = True
+        async with self._db.session() as session:
+            row = await session.get(InjectQueueItem, item.id)
+            if row is not None:
+                row.interrupted_turn = True
+                session.add(row)
+                await session.commit()
+
     async def interrupted(self) -> Sequence[InjectQueueItem]:
         async with self._db.session() as session:
             result = await session.exec(

src/beaver_gateway/storage/models.py

@@ -163,6 +163,9 @@ class InjectQueueItem(SQLModel, table=True):
     text: str
     status: str = Field(default="queued", index=True)
     turn_id: str | None = Field(default=None)
+    interrupted_turn: bool | None = Field(default=None)
+    """An urgent inject that cut a running turn; the header tells the model
+    the ``Request interrupted`` tool result above was this, not a refusal."""
     created_at: datetime = Field(default_factory=_utcnow)
     delivered_at: datetime | None = Field(default=None)
 

tests/test_conversations.py

@@ -274,10 +274,25 @@ async def test_urgent_interrupts_and_goes_first(world: World) -> None:
     ]
     assert client.prompts[0] == "first"
     assert client.prompts[1].startswith("[инжект: крон")
+    assert "прервал предыдущий тёрн" in client.prompts[1]
     assert client.prompts[1].endswith("ALERT")
     assert client.prompts[2] == "second"
 
 
+async def test_inject_header_is_configurable(world: World) -> None:
+    from beaver_gateway.core.conversations import ConversationTexts
+
+    world.conversations._texts = ConversationTexts(  # noqa: SLF001
+        inject_header=lambda ctx: (
+            f"[от {ctx.origin}, {ctx.priority}, cut={ctx.interrupted_turn}]"
+        )
+    )
+    conv = await world.conversations.create(kind="master", agent="a", origin="test")
+    await world.conversations.inject(conv, "hi", urgency="urgent", origin="panel")
+    await world.settle(conv, 1)
+    assert ScriptedClient.instances[0].prompts[0] == "[от panel, urgent, cut=False]\nhi"
+
+
 async def test_normal_injects_ride_with_the_next_user_message(world: World) -> None:
     conv = await world.conversations.create(kind="master", agent="a", origin="test")
     await world.conversations.inject(