参考文献模式合并为单一 referenceBatch 并统一连续编号
将同一轮检索结果合并为一个 referenceBatch,由系统统一从 [1] 连续编号,避免每个检索词各自重新编号;提示词引导优先并行检索后再作答。智能体列表接口并发加载技能等轻量信息以渲染卡片徽章,并新增参考文献相关测试。 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
1 parent
6792b4f4c4
commit
2c136c1fde
6 files changed
+164
-35
No files matched your search
+18
-16
@@ -703,12 +703,14 @@ def _get_cached_skills_prompt_section(
|
||||
|
||||
display_pattern = (
|
||||
"""
|
||||
**Skill Result Display Pattern:**
|
||||
- If a skill has `<display mode="citation">`, use the skill's returned result order as reference numbers. Cite facts from the first result as `[[1]]`, the second as `[[2]]`, and so on. Use only numbers that correspond to returned results.
|
||||
- If a skill has `<display mode="list">`, summarize the returned items as a concise list and keep item order stable.
|
||||
- If a skill has `<display mode="table">`, summarize the result and output `<skill-display tool="SKILL_OR_TOOL_NAME" mode="table" />` where the frontend should render the real table. Do not hand-write the table rows yourself.
|
||||
- If a skill has `<display mode="card">`, summarize the result and output `<skill-display tool="SKILL_OR_TOOL_NAME" mode="card" />` where the frontend should render the cards.
|
||||
- If a skill has `<display mode="custom">`, follow the custom display instruction in the `<display>` tag when presenting the skill result.
|
||||
**Skill Result Display Pattern:**
|
||||
- If RAG/reference display mode is enabled, do not number each skill result independently. Treat all retrieval results from the same user turn as one combined `referenceBatch` and cite with `[1]`, `[2]`, etc. from the system-provided reference prompt.
|
||||
- In RAG/reference display mode, do not output `<skill-display .../>`; the frontend renders cards, tables, lists, and citation details in the right reference panel.
|
||||
- If RAG/reference display mode is not enabled and a skill has `<display mode="citation">`, use the skill's returned result order as reference numbers. Cite facts from the first result as `[1]`, the second as `[2]`, and so on. Use only numbers that correspond to returned results.
|
||||
- If a skill has `<display mode="list">`, summarize the returned items as a concise list and keep item order stable.
|
||||
- If a skill has `<display mode="table">`, summarize the result and output `<skill-display tool="SKILL_OR_TOOL_NAME" mode="table" />` where the frontend should render the real table. Do not hand-write the table rows yourself.
|
||||
- If a skill has `<display mode="card">`, summarize the result and output `<skill-display tool="SKILL_OR_TOOL_NAME" mode="card" />` where the frontend should render the cards.
|
||||
- If a skill has `<display mode="custom">`, follow the custom display instruction in the `<display>` tag when presenting the skill result.
|
||||
"""
|
||||
if display_signature
|
||||
else ""
|
||||
@@ -993,16 +995,16 @@ Writing mode is enabled for this run.
|
||||
network_policy_section = _build_network_policy_section()
|
||||
rag_mode_section = (
|
||||
"""
|
||||
<rag_citation_mode>
|
||||
RAG/reference display mode is enabled.
|
||||
When a search, knowledge, skill, or MCP tool returns a result list:
|
||||
- If the skill has a display adapter, follow that adapter first.
|
||||
- For `citation`, cite returned items with `[[1]]`, `[[2]]`, etc. Only cite real returned items.
|
||||
- For `table`, do not hand-write the table. Summarize and place `<skill-display tool="SKILL_OR_TOOL_NAME" mode="table" />` where the table should appear. The frontend will render the real tool result as a stable table.
|
||||
- For `list` or `card`, summarize and place `<skill-display tool="SKILL_OR_TOOL_NAME" mode="list" />` or `<skill-display tool="SKILL_OR_TOOL_NAME" mode="card" />` where the structured result should appear.
|
||||
- If no display adapter exists, default to citation mode with numbered citations.
|
||||
</rag_citation_mode>
|
||||
"""
|
||||
<rag_citation_mode>
|
||||
RAG/reference display mode is enabled.
|
||||
When a search, knowledge, skill, or MCP tool returns a result list:
|
||||
- If multiple search terms are needed, prefer calling the retrieval tools for all known terms in the same step, then answer after the tool results are returned.
|
||||
- Treat every retrieval result from the same user turn as one combined `referenceBatch`; do not restart numbering for each search term, tool call, skill, or display adapter.
|
||||
- Use only the system-provided merged reference numbers in the final answer, formatted as `[1]`, `[2]`, `[1][3][5]`.
|
||||
- Do not output `<skill-display .../>` in RAG/reference mode. The frontend renders configured cards, tables, lists, and citation details in the right reference panel.
|
||||
- If no display adapter exists, the frontend defaults the right reference panel to numbered citations.
|
||||
</rag_citation_mode>
|
||||
"""
|
||||
if rag_mode_enabled
|
||||
else ""
|
||||
)
|
||||
|
||||
+4
-2
@@ -80,13 +80,15 @@ class RagCitationMiddleware(AgentMiddleware[AgentState]):
|
||||
return None
|
||||
summary = build_reference_debug_summary(messages, [batch] if batch else None)
|
||||
_log_rag_debug(
|
||||
"middleware reference-prompt enabled=true prompt=true messages=%s tool_messages=%s "
|
||||
"raw_items=%s normalized_items=%s kept_items=%s latest_batch_sources=%s",
|
||||
"middleware reference-prompt enabled=true prompt=true single_batch=true messages=%s "
|
||||
"tool_messages=%s raw_items=%s normalized_items=%s kept_items=%s latest_batch_id=%s "
|
||||
"latest_batch_sources=%s",
|
||||
summary["messages"],
|
||||
summary["tool_messages"],
|
||||
summary["raw_items"],
|
||||
summary["normalized_items"],
|
||||
summary["kept_items"],
|
||||
(batch or {}).get("id"),
|
||||
len((batch or {}).get("sources") or []),
|
||||
)
|
||||
_log_rag_debug(
|
||||
|
||||
@@ -37,24 +37,45 @@ def build_reference_batches(messages: list[Any]) -> list[dict[str, Any]]:
|
||||
return batches
|
||||
|
||||
|
||||
def build_latest_reference_prompt(messages: list[Any]) -> tuple[str, dict[str, Any] | None]:
|
||||
def build_latest_reference_batch(messages: list[Any]) -> dict[str, Any] | None:
|
||||
"""Return the single combined citation batch for the latest human turn."""
|
||||
human_index = _find_latest_human_index(messages)
|
||||
if human_index is None:
|
||||
return "", None
|
||||
return None
|
||||
sources = _collect_sources(messages[human_index + 1 :])
|
||||
if not sources:
|
||||
return "", None
|
||||
return None
|
||||
assistant = _find_reference_batch_anchor(messages, human_index + 1, len(messages))
|
||||
message_id = _message_id(assistant) or _message_id(messages[human_index]) or str(human_index)
|
||||
batch = {
|
||||
return {
|
||||
"id": message_id,
|
||||
"assistant_message_id": message_id,
|
||||
"source_count": len(sources),
|
||||
"sources": sources,
|
||||
}
|
||||
|
||||
|
||||
def build_latest_reference_prompt(messages: list[Any]) -> tuple[str, dict[str, Any] | None]:
|
||||
batch = build_latest_reference_batch(messages)
|
||||
if batch is None:
|
||||
return "", None
|
||||
return build_reference_prompt_for_batch(batch), batch
|
||||
|
||||
|
||||
def build_reference_prompt_for_batch(batch: dict[str, Any]) -> str:
|
||||
"""Build the authoritative model prompt from one already-combined batch."""
|
||||
sources = [
|
||||
source
|
||||
for source in batch.get("sources", [])
|
||||
if isinstance(source, dict)
|
||||
]
|
||||
source_count = len(sources)
|
||||
lines = [
|
||||
"本轮检索结果已由系统统一合并并重新编号。最终回答必须使用下面这套编号。",
|
||||
"如果你仍需要补充检索,请先继续调用工具;最终回答只能引用最新合并后的编号。",
|
||||
f"本轮所有检索结果已合并为同一个 referenceBatch,共 {source_count} 条,编号从 [1] 到 [{source_count}] 唯一连续。",
|
||||
"下面这一批就是当前回答唯一可用的参考文献批次;不要按检索词、工具调用次数或工具原始编号重新分批编号。",
|
||||
"如果你仍需要补充检索,请只继续调用工具,不要输出正文;最终回答只能引用补充检索后系统重新合并出的最新 referenceBatch。",
|
||||
"",
|
||||
f"<reference_batch source_count=\"{source_count}\">",
|
||||
]
|
||||
for source in sources:
|
||||
title = source.get("title") or f"参考文献 {source.get('index')}"
|
||||
@@ -73,18 +94,20 @@ def build_latest_reference_prompt(messages: list[Any]) -> tuple[str, dict[str, A
|
||||
if content:
|
||||
lines.append(content)
|
||||
lines.append("")
|
||||
lines.append("</reference_batch>")
|
||||
lines.append("")
|
||||
lines.extend(
|
||||
[
|
||||
"引用规则:",
|
||||
"1. 回答正文中的关键结论必须标注对应编号,例如 [1]、[2]。",
|
||||
"2. 多个来源可写为 [1][3][5]。",
|
||||
"3. 只能引用上方存在的编号,不要编造编号。",
|
||||
"4. 不要使用工具原始编号,也不要重新排序编号。",
|
||||
"4. 不要使用工具原始编号,也不要重新排序、重新分批或重新从 [1] 开始编号。",
|
||||
"5. 不要在正文末尾堆完整参考文献列表,前端会展示右侧参考文献。",
|
||||
"6. 参考文献模式下不要输出 <skill-display .../> 占位符,也不要在正文里额外渲染卡片、表格或列表;前端会在右侧参考文献区域展示。",
|
||||
]
|
||||
)
|
||||
return "\n".join(lines).strip(), batch
|
||||
return "\n".join(lines).strip()
|
||||
|
||||
|
||||
def _find_latest_human_index(messages: list[Any]) -> int | None:
|
||||
|
||||
Reference in new issue
Block a user