feat(agents,backends,conversations,frontends,ui): progress thinking is told apart by its agent, raycast reasoning stays collapsed, read_conversation keeps progress

tests/test_markdown_renderer.py

@@ -1,7 +1,11 @@
+from pathlib import Path
+
 from anthropic.types import Message, TextBlock, ThinkingBlock, ToolUseBlock, Usage
 
 from beaver_gateway.frontends.markdown import history, parser, renderer
+from beaver_gateway.frontends.markdown.crossfront import CrossFrontendLogger
 from beaver_gateway.frontends.markdown.parser import TextSegment, ToolSegment
+from beaver_gateway.frontends.turn_record import TurnRecord
 
 
 def _message(*blocks: object) -> Message:
@@ -25,9 +29,10 @@ def test_tool_calls_leave_no_blank_lines() -> None:
         ToolUseBlock(type="tool_use", id="t3", name="Grep", input={"pattern": "y"}),
         TextBlock(type="text", text="second"),
     )
-    assert renderer.render_assistant_message(message) == (
-        "### Assistant:\n\nfirst\n\nsecond\n"
-    )
+    for progress in (True, False):
+        assert renderer.render_assistant_message(
+            message, thinking_is_progress=progress
+        ) == ("### Assistant:\n\nfirst\n\nsecond\n")
 
 
 def _progress_turn() -> Message:
@@ -44,38 +49,88 @@ def _progress_turn() -> Message:
     )
 
 
+def _reasoning_turn() -> Message:
+    return _message(
+        ThinkingBlock(type="thinking", thinking="one", signature=""),
+        ToolUseBlock(type="tool_use", id="t1", name="web_search", input={}),
+        ThinkingBlock(type="thinking", thinking="two", signature=""),
+        ToolUseBlock(type="tool_use", id="t2", name="read_page", input={}),
+        ThinkingBlock(type="thinking", thinking="three", signature=""),
+        TextBlock(type="text", text="answer"),
+        ThinkingBlock(type="thinking", thinking="four\nfive", signature=""),
+        TextBlock(type="text", text="more"),
+    )
+
+
+REASONING_TURN_AS_BEFORE = (
+    "### Assistant:\n\n> [!thinking]-\n> one\n\nanswer\n\n"
+    "> [!thinking]-\n> four\n> five\n\nmore\n"
+)
+
+
+def _full_circle(
+    message: Message, *, thinking_is_progress: bool
+) -> history.ForkOutcome:
+    body = renderer.render_user_text("status?") + renderer.render_assistant_message(
+        message, thinking_is_progress=thinking_is_progress
+    )
+    turns = parser.parse(body + renderer.render_user_text("thanks")).turns
+    stored = [
+        {"role": "user", "content": "status?"},
+        {"role": "assistant", "content": [b.model_dump() for b in message.content]},
+    ]
+    outcome = history.diff_and_fork(
+        stored=stored, incoming=turns, thinking_is_progress=thinking_is_progress
+    )
+    assert outcome.divergence_index is None
+    assert not outcome.edited
+    assert outcome.messages[1] == stored[1]
+    return outcome
+
+
 def test_progress_renders_as_text_in_place_and_empty_thinking_vanishes() -> None:
-    assert renderer.render_assistant_message(_progress_turn()) == (
+    assert renderer.render_assistant_message(
+        _progress_turn(), thinking_is_progress=True
+    ) == (
         "### Assistant:\n\nChecking the containers.\n\nAll up, reading logs.\n\n"
         "Found it.\nOne more.\n\nThe stack is healthy.\n"
     )
 
 
+def test_reasoning_renders_as_collapsed_callouts_as_before() -> None:
+    assert (
+        renderer.render_assistant_message(_reasoning_turn(), thinking_is_progress=False)
+        == REASONING_TURN_AS_BEFORE
+    )
+
+
 def test_progress_round_trips_through_the_parser_and_the_stored_history() -> None:
     message = _progress_turn()
-    body = renderer.render_user_text("status?") + renderer.render_assistant_message(
-        message
-    )
-    turns = parser.parse(body + renderer.render_user_text("thanks")).turns
-    assert turns[1].structure == (
+    body = renderer.render_assistant_message(message, thinking_is_progress=True)
+    assert parser.parse_assistant_structure(body.removeprefix("### Assistant:")) == [
         TextSegment(
             text="Checking the containers.\n\nAll up, reading logs.\n\n"
             "Found it.\nOne more.\n\nThe stack is healthy."
-        ),
-    )
-    stored = [
-        {"role": "user", "content": "status?"},
-        {"role": "assistant", "content": [b.model_dump() for b in message.content]},
+        )
     ]
-    outcome = history.diff_and_fork(stored=stored, incoming=turns)
-    assert outcome.divergence_index is None
-    assert not outcome.edited
-    assert outcome.messages[1] == stored[1]
+    _full_circle(message, thinking_is_progress=True)
+
+
+def test_reasoning_round_trips_through_the_parser_and_the_stored_history() -> None:
+    message = _reasoning_turn()
+    body = renderer.render_assistant_message(message, thinking_is_progress=False)
+    assert parser.parse_assistant_structure(body.removeprefix("### Assistant:")) == [
+        TextSegment(text="answer"),
+        TextSegment(text="more"),
+    ]
+    _full_circle(message, thinking_is_progress=False)
 
 
 def test_an_edited_progress_turn_rebuilds_instead_of_rewriting_signed_blocks() -> None:
     message = _progress_turn()
-    body = renderer.render_assistant_message(message).replace("Found it", "Fixed it")
+    body = renderer.render_assistant_message(
+        message, thinking_is_progress=True
+    ).replace("Found it", "Fixed it")
     turns = parser.parse(
         renderer.render_user_text("status?") + body + renderer.render_user_text("ok")
     ).turns
@@ -83,7 +138,9 @@ def test_an_edited_progress_turn_rebuilds_instead_of_rewriting_signed_blocks() -
         {"role": "user", "content": "status?"},
         {"role": "assistant", "content": [b.model_dump() for b in message.content]},
     ]
-    outcome = history.diff_and_fork(stored=stored, incoming=turns)
+    outcome = history.diff_and_fork(
+        stored=stored, incoming=turns, thinking_is_progress=True
+    )
     assert outcome.edited
     assert outcome.messages[1] == {
         "role": "assistant",
@@ -91,6 +148,22 @@ def test_an_edited_progress_turn_rebuilds_instead_of_rewriting_signed_blocks() -
     }
 
 
+async def test_a_raycast_archive_with_reasoning_renders_as_before(
+    tmp_path: Path,
+) -> None:
+    logger = CrossFrontendLogger(vault_path=tmp_path, logged_subdir="logged")
+    record = TurnRecord(
+        agent_name="beaver-gemini-pro-high",
+        input_messages=[{"role": "user", "content": "q"}],
+        output_message=_reasoning_turn(),
+        thinking_is_progress=False,
+    )
+    await logger.handle(record)
+    [written] = (tmp_path / "logged" / "beaver-gemini-pro-high").glob("*.md")
+    body = parser.parse(written.read_text()).body
+    assert f"---\n\n{REASONING_TURN_AS_BEFORE}\n---\n\n### User:" in body
+
+
 def test_a_file_with_collapsed_thinking_callouts_parses_as_before() -> None:
     body = (
         "### User:\n\nq\n\n---\n\n### Assistant:\n\n"