fix(bark): refine compact briefing template

tests/test_bark_publisher.py

@@ -257,16 +257,23 @@ def handler(request: httpx.Request) -> httpx.Response:
     assert "Private body" not in caplog.text
 
 
-def test_split_plain_message_prefers_line_boundary() -> None:
-    assert split_plain_message("first line\nsecond line", 12) == ("first line", "\nsecond line")
+def test_split_plain_message_consumes_line_boundary() -> None:
+    chunks = split_plain_message("first line\nsecond line", 12)
 
+    assert chunks == ("first line", "second line")
+    assert all(not chunk.startswith("\n") and not chunk.endswith("\n") for chunk in chunks)
 
-def test_split_plain_message_uses_the_minimum_number_of_chunks() -> None:
+
+def test_split_plain_message_omits_empty_chunk_after_final_boundary() -> None:
+    chunks = split_plain_message("x" * 12 + "\n", 12)
+
+    assert chunks == ("x" * 12,)
+
+
+def test_split_plain_message_preserves_non_boundary_newlines() -> None:
     chunks = split_plain_message("x" * 100 + "\n" + "y" * 1199, 650)
 
-    assert len(chunks) == 2
-    assert all(len(chunk) <= 650 for chunk in chunks)
-    assert "".join(chunks) == "x" * 100 + "\n" + "y" * 1199
+    assert chunks == ("x" * 100 + "\n" + "y" * 549, "y" * 650)
 
 
 @pytest.mark.parametrize("limit", (0, -1))

tests/test_prompts.py

@@ -63,6 +63,8 @@ def test_prompt_uses_a_soft_briefing_target_and_hard_output_limits() -> None:
 
 def test_prompt_requires_attribution_and_preserves_source_conflicts() -> None:
     assert "headline_source_ids 以及 conclusions" in SYSTEM_PROMPT
+    assert "只能包含纯文本" in SYSTEM_PROMPT
+    assert "不得使用 Markdown" in SYSTEM_PROMPT
     assert "不得拼接成无争议的单一结论" in SYSTEM_PROMPT
     assert "优先采用可识别的当地权威气象机构" in SYSTEM_PROMPT
     assert "input.required_advice_topics" in SYSTEM_PROMPT

tests/test_render.py

@@ -81,9 +81,48 @@ def test_bark_text_renderer_uses_numbered_sources_without_urls() -> None:
 
     rendered = BarkTextRenderer().render_briefing(result, (article,), (context,))
 
-    assert rendered.body == "Daily [1]\n- Rain [1][2]\nSources: [1] Feed; [2] Weather API"
+    assert rendered.body == "Daily [1]\nRain [1][2]\n[1] Feed\n[2] Weather API"
     assert article.url not in rendered.body
     assert context.url not in rendered.body
+    assert "- " not in rendered.body
+
+
+def test_bark_text_renderer_merges_source_ids_with_the_same_display_name() -> None:
+    weather = SourceDocument("weather:open-meteo", "Open-Meteo", "https://example.invalid/weather", "Forecast")
+    air_quality = SourceDocument(
+        "air-quality:open-meteo",
+        " open-meteo ",
+        "https://example.invalid/air-quality",
+        "Air quality",
+    )
+    result = BriefingResult(
+        "Daily",
+        ("weather:open-meteo",),
+        (Conclusion("Rain", ("weather:open-meteo", "air-quality:open-meteo")),),
+        advice=(Advice(AdviceTopic.MASK, "Limit exposure", ("air-quality:open-meteo",)),),
+        output_language="en",
+    )
+
+    rendered = BarkTextRenderer().render_briefing(result, (), (weather, air_quality))
+
+    assert rendered.body == "Daily [1]\nRain [1]\nAdvice\nLimit exposure [1]\n[1] Open-Meteo"
+    assert "[2]" not in rendered.body
+    assert "Sources:" not in rendered.body
+
+
+def test_bark_text_renderer_uses_source_ids_for_distinct_blank_names() -> None:
+    weather = SourceDocument("weather:blank", " ", "https://example.invalid/weather", "Forecast")
+    air_quality = SourceDocument("air-quality:blank", "", "https://example.invalid/air-quality", "Air quality")
+    result = BriefingResult(
+        "Daily",
+        ("weather:blank",),
+        (Conclusion("Rain", ("weather:blank", "air-quality:blank")),),
+        output_language="en",
+    )
+
+    rendered = BarkTextRenderer().render_briefing(result, (), (weather, air_quality))
+
+    assert rendered.body == ("Daily [1]\nRain [1][2]\n[1] weather:blank\n[2] air-quality:blank")
 
 
 def test_bark_text_renderer_trims_outer_whitespace() -> None:
@@ -92,7 +131,7 @@ def test_bark_text_renderer_trims_outer_whitespace() -> None:
 
     rendered = BarkTextRenderer().render_briefing(result, (), (context,))
 
-    assert rendered.body == "Daily [1]\nSources: [1] Weather API"
+    assert rendered.body == "Daily [1]\n[1] Weather API"
     assert rendered.visible_length == len(rendered.body)
 
 
@@ -114,12 +153,12 @@ def test_bark_text_renderer_compacts_warning_disaster_and_advice_sections() -> N
     assert rendered.body == (
         "Rain today [1]\n"
         "Weather warnings\n"
-        "- Heavy rain (active): Avoid low areas [1]\n"
+        "Heavy rain (active): Avoid low areas [1]\n"
         "Natural disaster updates\n"
-        "- Storm approaching [1]\n"
+        "Storm approaching [1]\n"
         "Advice\n"
-        "- Exercise indoors [1]\n"
-        "Sources: [1] Weather API"
+        "Exercise indoors [1]\n"
+        "[1] Weather API"
     )
 
 

weather_briefing/data/system_prompt.txt

@@ -16,6 +16,8 @@ recent_context_documents 中的 language 是来源正文的实际语言；来源
 不得把翻译结果冒充来源原文或改变专名、预警编号和数值。
 headline_source_ids 以及 conclusions、active_warnings、disaster_tracking 和 advice
 中的每一项都必须包含至少一个 source_id。
+headline、conclusions[].text、active_warnings 中的 title、status、detail、disaster_tracking[].text
+以及 advice[].text 只能包含纯文本，不得使用 Markdown 标题、列表、强调、代码或链接语法；章节标题和项目符号由发布端统一渲染。
 headline 必须是一句简洁、信息密集的标题，将当下最重要的天气概况浓缩其中；优先包含天气现象、
 高低温或显著体感，以及需要立即准备的短时降水等变化，不要另写摘要段落。
 不同来源对同一时段的天气现象有冲突时，不得拼接成无争议的单一结论；应明确说明差异，

weather_briefing/delivery/bark.py

@@ -174,7 +174,7 @@ def bark_error_reason(response: httpx.Response) -> tuple[str, bool]:
 
 
 def split_plain_message(body: str, limit: int) -> tuple[str, ...]:
-    """Split plain text at line boundaries when possible."""
+    """Split into display-ready chunks, consuming newlines used as boundaries."""
     if limit <= 0:
         raise ValueError("Message split limit must be positive")
     if not body or len(body) <= limit:
@@ -184,12 +184,15 @@ def split_plain_message(body: str, limit: int) -> tuple[str, ...]:
     while len(remaining) > limit:
         remaining_chunk_count = math.ceil(len(remaining) / limit)
         earliest_split = len(remaining) - (remaining_chunk_count - 1) * limit
-        split_at = remaining.rfind("\n", earliest_split, limit + 1)
-        if split_at < earliest_split:
-            split_at = limit
-        chunks.append(remaining[:split_at])
-        remaining = remaining[split_at:]
-    chunks.append(remaining)
+        newline_at = remaining.rfind("\n", earliest_split, limit + 1)
+        if newline_at >= earliest_split:
+            chunks.append(remaining[:newline_at])
+            remaining = remaining[newline_at + 1 :]
+        else:
+            chunks.append(remaining[:limit])
+            remaining = remaining[limit:]
+    if remaining:
+        chunks.append(remaining)
     return tuple(chunks)
 
 

weather_briefing/delivery/renderers.py

@@ -218,7 +218,7 @@ def render_briefing(
         source_references.update(
             {document.id: self._source_reference(document.name, document.url) for document in context}
         )
-        numbered_references, source_footer = _numbered_source_references(result, source_references, labels)
+        numbered_references, source_footer = _bark_numbered_source_references(result, source_references)
         lines = [
             f"{result.headline} "
             f"{_plain_attribution(result.headline_source_ids, numbered_references, labels, numbered=True)}"
@@ -227,7 +227,7 @@ def render_briefing(
         if result.active_warnings:
             lines.append(labels["warnings"])
             lines.extend(
-                f"- {warning.title}{labels['status_open']}{warning.status}{labels['status_close']}"
+                f"{warning.title}{labels['status_open']}{warning.status}{labels['status_close']}"
                 f"{labels['detail_separator']}{warning.detail} "
                 f"{_plain_attribution(warning.source_ids, numbered_references, labels, numbered=True)}"
                 for warning in result.active_warnings
@@ -295,8 +295,7 @@ def _compact_plain_items(
         return []
     lines = [title] if title is not None else []
     lines.extend(
-        f"- {item.text} {_plain_attribution(item.source_ids, source_references, labels, numbered=True)}"
-        for item in items
+        f"{item.text} {_plain_attribution(item.source_ids, source_references, labels, numbered=True)}" for item in items
     )
     return lines
 
@@ -329,11 +328,7 @@ def _numbered_source_references(
     source_references: dict[str, str],
     labels: Mapping[str, str],
 ) -> tuple[dict[str, str], str]:
-    ordered_source_ids = list(result.headline_source_ids)
-    for items in (result.conclusions, result.active_warnings, result.disaster_tracking, result.advice):
-        for item in items:
-            ordered_source_ids.extend(item.source_ids)
-    ordered_source_ids = list(dict.fromkeys(ordered_source_ids))
+    ordered_source_ids = _ordered_source_ids(result)
     numbered_references = {source_id: f"[{index}]" for index, source_id in enumerate(ordered_source_ids, start=1)}
     source_list = labels["plain_source_separator"].join(
         f"{numbered_references[source_id]} {source_references[source_id]}" for source_id in ordered_source_ids
@@ -342,6 +337,33 @@ def _numbered_source_references(
     return numbered_references, footer
 
 
+def _bark_numbered_source_references(
+    result: BriefingResult,
+    source_references: dict[str, str],
+) -> tuple[dict[str, str], str]:
+    numbered_references: dict[str, str] = {}
+    numbers_by_name: dict[str, str] = {}
+    source_lines: list[str] = []
+    for source_id in _ordered_source_ids(result):
+        source_name = " ".join(source_references[source_id].split()) or source_id
+        normalized_name = source_name.casefold()
+        number = numbers_by_name.get(normalized_name)
+        if number is None:
+            number = f"[{len(numbers_by_name) + 1}]"
+            numbers_by_name[normalized_name] = number
+            source_lines.append(f"{number} {source_name}")
+        numbered_references[source_id] = number
+    return numbered_references, "\n".join(source_lines)
+
+
+def _ordered_source_ids(result: BriefingResult) -> list[str]:
+    ordered_source_ids = list(result.headline_source_ids)
+    for items in (result.conclusions, result.active_warnings, result.disaster_tracking, result.advice):
+        for item in items:
+            ordered_source_ids.extend(item.source_ids)
+    return list(dict.fromkeys(ordered_source_ids))
+
+
 def _briefing_labels(language: str) -> Mapping[str, str]:
     selected = _BRIEFING_LANGUAGE_SUPPORT.match(language)
     return _BRIEFING_LABELS[selected]