feat(frontends,ui): opus 5.5 progress updates show in the web panel and the vault transcript

src/beaver_gateway/frontends/markdown/history.py

@@ -255,8 +255,9 @@ def _summarize_assistant_group(
 ) -> tuple[str, list[str], int]:
     """Compute (spoken_text, tool_skeleton, text_segment_count) for a display group.
 
-    Must mirror ``parser.parse_assistant_structure``: text blocks collapse into
-    one segment, tool_use becomes a skeleton entry, tool_result/thinking are invisible.
+    Must mirror what the renderer writes and the parser reads back: text and
+    non-empty thinking collapse into one segment, tool_use becomes a skeleton
+    entry, tool_result and empty thinking are invisible.
     """
     from beaver_gateway.frontends.markdown.parser import TextSegment, ToolSegment
 
@@ -282,12 +283,10 @@ def _summarize_assistant_group(
         for blk in content:
             if not isinstance(blk, dict):
                 continue
-            btype = blk.get("type")
-            if btype == "text":
-                text = str(blk.get("text", "")).strip()
-                if text:
-                    pending.append(text)
-            elif btype == "tool_use":
+            text = _spoken(blk)
+            if text:
+                pending.append(text)
+            elif blk.get("type") == "tool_use":
                 _flush()
                 segments.append(ToolSegment(name=str(blk.get("name", ""))))
     _flush()
@@ -440,11 +439,16 @@ def _splice_assistant_group(
     return _splice_by_rebuild(stored_group=stored_group, incoming=incoming)
 
 
-def _is_text_block(blk: object) -> bool:
+def _spoken(blk: object) -> str:
     if not isinstance(blk, dict):
-        return False
+        return ""
     block = cast("dict[str, Any]", blk)
-    return block.get("type") == "text" and bool(str(block.get("text", "")).strip())
+    kind = block.get("type")
+    if kind == "text":
+        return str(block.get("text", "")).strip()
+    if kind == "thinking":
+        return str(block.get("thinking", "")).strip()
+    return ""
 
 
 def _splice_in_place(
@@ -454,7 +458,8 @@ def _splice_in_place(
 
     Preserves message/thinking-block boundaries that a rebuild would lose
     (and that a cache-hit continuation depends on); returns ``None`` if
-    text blocks aren't 1:1 with incoming segments, so the caller rebuilds instead.
+    text blocks aren't 1:1 with incoming segments or the prose runs through
+    signed thinking blocks, so the caller rebuilds instead.
     """
     from beaver_gateway.frontends.markdown.parser import TextSegment
 
@@ -466,9 +471,12 @@ def _splice_in_place(
         content = msg.get("content")
         if msg["role"] != "assistant" or not isinstance(content, list):
             continue
-        positions.extend(
-            (mi, bi) for bi, blk in enumerate(content) if _is_text_block(blk)
-        )
+        for bi, blk in enumerate(cast("list[dict[str, Any]]", content)):
+            if not _spoken(blk):
+                continue
+            if blk["type"] == "thinking":
+                return None
+            positions.append((mi, bi))
     if len(positions) != len(new_texts):
         return None
 

src/beaver_gateway/frontends/markdown/renderer.py

@@ -1,7 +1,7 @@
 """Render Anthropic ``Message`` (and user turns) into markdown.
 
 One-way: produces the human-facing artifact only. The parser strips
-tool/thinking callouts separately when reshaping history for backend replay.
+the tool/thinking callouts of older files when reshaping history for replay.
 """
 
 from __future__ import annotations
@@ -12,8 +12,6 @@ from typing import TYPE_CHECKING, Any, cast
 from anthropic.types import Message, TextBlock, ThinkingBlock
 
 if TYPE_CHECKING:
-    from collections.abc import Iterable
-
     from anthropic.types import MessageParam
 
 
@@ -48,21 +46,14 @@ def render_assistant_text(text: str) -> str:
 def render_assistant_message(message: Message) -> str:
     """Render an assistant ``Message`` into a turn block.
 
-    Tool-use blocks render to nothing; of the thinking blocks between two
-    texts only the first one reaches the file.
+    Tool-use blocks and empty thinking render to nothing; a non-empty thinking
+    block is what the model wrote between tool calls and renders as text.
     """
     parts: list[str] = ["### Assistant:", ""]
-    after_thinking = False
     for block in message.content:
-        thinking = isinstance(block, ThinkingBlock)
-        if thinking and after_thinking:
-            continue
-        lines = list(_render_block(block))
-        if not lines:
-            continue
-        after_thinking = thinking
-        parts.extend(lines)
-        parts.append("")
+        text = _spoken(block)
+        if text:
+            parts.extend((text, ""))
     return "\n".join(parts).rstrip() + "\n"
 
 
@@ -108,21 +99,12 @@ def summarize_tool_input(name: str, tool_input: object) -> str:
     return name
 
 
-def _render_block(block: object) -> Iterable[str]:
+def _spoken(block: object) -> str:
     if isinstance(block, TextBlock):
-        text = (block.text or "").strip()
-        if text:
-            yield text
-        return
+        return (block.text or "").strip()
     if isinstance(block, ThinkingBlock):
-        yield from _render_thinking(block.thinking or "")
-        return
-
-
-def _render_thinking(text: str) -> Iterable[str]:
-    yield "> [!thinking]-"
-    for line in text.strip().splitlines() or [""]:
-        yield f"> {line}" if line else ">"
+        return (block.thinking or "").strip()
+    return ""
 
 
 def adaptive_fence(content: str) -> str: