feat(telegram): stop generation from the client, questions survive a restart, images reach the model inline, albums become one turn

src/beaver_gateway/backends/claude_sdk.py

@@ -313,7 +313,7 @@ class ClaudeSdkBackend:
         if not history or history[-1].get("role") != "user":
             msg = "the last message must be a user turn"
             raise ValueError(msg)
-        prompt = _prompt_text(history[-1].get("content"))
+        prompt = _prompt(history[-1].get("content"))
         prior = history[:-1]
         key = conversation_id or fingerprint(prior)
         spec = _SessionSpec(kind=kind, pinned=pinned, tools=tools)
@@ -404,7 +404,7 @@ class ClaudeSdkBackend:
     async def _run_turn(
         self,
         live: Session,
-        prompt: str,
+        prompt: str | list[dict[str, Any]],
         turn: _Turn,
         observer: Callable[[Any], None] | None = None,
         capture: TurnCapture | None = None,
@@ -421,7 +421,7 @@ class ClaudeSdkBackend:
         raw: list[Any] = []
         next_index = 0
         offset = 0
-        await live.client.query(prompt)
+        await live.client.query(prompt if isinstance(prompt, str) else _stream(prompt))
         async for message in live.client.receive_response():
             if observer is not None:
                 observer(message)
@@ -845,14 +845,27 @@ def _mcp_disallowed(
     return out
 
 
-def _prompt_text(content: Any) -> str:
+def _prompt(content: Any) -> str | list[dict[str, Any]]:
+    """Plain text when the turn is text only; the block list when images ride along."""
     text = text_of(content)
     if not text:
         msg = "user message has no text content"
         raise ValueError(msg)
+    if isinstance(content, list) and any(
+        isinstance(b, dict) and b.get("type") != "text" for b in content
+    ):
+        return [dict(b) for b in content if isinstance(b, dict)]
     return text
 
 
+async def _stream(content: list[dict[str, Any]]) -> AsyncIterator[dict[str, Any]]:
+    yield {
+        "type": "user",
+        "message": {"role": "user", "content": content},
+        "parent_tool_use_id": None,
+    }
+
+
 def synthesize_turn_messages(raw: Iterable[Any]) -> list[dict[str, Any]]:
     out: list[dict[str, Any]] = []
     for message in raw:

src/beaver_gateway/backends/sessions.py

@@ -20,7 +20,7 @@ from typing import TYPE_CHECKING, Any, Protocol
 import psutil
 
 if TYPE_CHECKING:
-    from collections.abc import AsyncIterator, Iterator, Mapping
+    from collections.abc import AsyncIterable, AsyncIterator, Iterator, Mapping
 
 __all__ = ["DEFAULT_TTL", "Session", "SessionClient", "SessionPool", "cgroup_limit"]
 
@@ -41,7 +41,7 @@ _RSS_HEADROOM = 0.8
 
 class SessionClient(Protocol):
     async def connect(self) -> None: ...
-    async def query(self, prompt: str) -> None: ...
+    async def query(self, prompt: str | AsyncIterable[dict[str, Any]]) -> None: ...
     def receive_response(self) -> AsyncIterator[Any]: ...
     async def interrupt(self) -> None: ...
     async def disconnect(self) -> None: ...