fix(writer): 章节级数据注入 + 修复 prompt 从未渲染的关键缺陷
- models: CHAPTER_SHEET_TYPES/CHAPTER_IMPACT_ELEMENT 章节→数据映射（design §6.8 ①）
- models: _format_chapter_data 按章定向渲染 ExcelTable 为 Markdown；GENERIC 自由記述作通用背景
- models: _format_impact 支持按 ElementType 过滤（章节级影响上下文）
- writer_agent: {{source}}(全量repr) → {{data}}(章节数据)；新增【主题约束】
- writer_agent: 关键修复——真实 PromptRegistry.get 返回模板字符串，
引擎对 str 不做变量渲染，LLM 实际收到的是 {{占位符}} 原文；
_resolve_prompt 统一包装为 Prompt 对象保证渲染
- 真实试运行验证：13 章主题全部正确、引用影响调查数据、语言漂移消除
This commit is contained in:
@@ -4,6 +4,37 @@ from __future__ import annotations
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Literal
|
||||
|
||||
from genesis.data_models import ElementType, SheetType
|
||||
|
||||
# 单表渲染行数上限（防 token 爆炸；样本量小，通常不触发）
|
||||
MAX_ROWS_PER_TABLE = 200
|
||||
|
||||
# design.md §6.8 ①「DataGate.load(structured_source, selector=该章数据)」的章节级选择器：
|
||||
# 章节占位符 id → 本章对应的 Excel Sheet 类型
|
||||
# - introduction 概览章注入全部类型表
|
||||
# - 未登记的章节 id 缺省为空列表 → 仅 GENERIC 自由記述作背景
|
||||
_ALL_SHEET_TYPES = list(SheetType)
|
||||
CHAPTER_SHEET_TYPES: dict[str, list[SheetType]] = {
|
||||
"introduction": _ALL_SHEET_TYPES,
|
||||
"function_list": [SheetType.FUNCTION],
|
||||
"screen_list": [SheetType.SCREEN],
|
||||
"report_list": [SheetType.REPORT],
|
||||
"db_design": [SheetType.DATABASE],
|
||||
"if_definition": [SheetType.INTERFACE],
|
||||
"batch_list": [SheetType.BATCH],
|
||||
}
|
||||
|
||||
# 章节 id → 影响调查要素类型（None = 展示全部，用于概览章）
|
||||
CHAPTER_IMPACT_ELEMENT: dict[str, ElementType | None] = {
|
||||
"function_list": ElementType.FUNCTION,
|
||||
"screen_list": ElementType.SCREEN,
|
||||
"report_list": ElementType.REPORT,
|
||||
"db_design": ElementType.DB,
|
||||
"if_definition": ElementType.IF,
|
||||
"batch_list": ElementType.BATCH,
|
||||
"introduction": None,
|
||||
}
|
||||
|
||||
|
||||
@dataclass
|
||||
class ContentBlock:
|
||||
@@ -90,18 +121,73 @@ class GenerationContext:
|
||||
"design_rules": "\n".join(self.design_rules),
|
||||
"template_styles": ", ".join(sorted(self.template_styles)),
|
||||
"prior_state": str(self.prior_state) if self.prior_state is not None else "",
|
||||
"source": str(self.structured_source) if self.structured_source is not None else "",
|
||||
"impact": _format_impact(self.impact_report),
|
||||
"data": _format_chapter_data(
|
||||
self.structured_source,
|
||||
CHAPTER_SHEET_TYPES.get(self.chapter_id, []),
|
||||
),
|
||||
"impact": _format_impact(
|
||||
self.impact_report,
|
||||
CHAPTER_IMPACT_ELEMENT.get(self.chapter_id, None),
|
||||
),
|
||||
}
|
||||
|
||||
|
||||
def _format_impact(report: object | None) -> str:
|
||||
"""将影响调查书格式化为 prompt 可读文本（无报告/无分析时为空串）。"""
|
||||
def _render_table(tb) -> list[str]:
|
||||
"""将单张 ExcelTable 渲染为可读 Markdown 行（表名行 + 表头 + 分隔 + 数据行）。"""
|
||||
lines = [f"### 表: {tb.name}（{tb.detected_type.value}）"]
|
||||
headers = list(tb.headers)
|
||||
lines.append("| " + " | ".join(headers) + " |")
|
||||
lines.append("|" + "|".join([" --- "] * len(headers)) + "|")
|
||||
for row in tb.rows[:MAX_ROWS_PER_TABLE]:
|
||||
cells = []
|
||||
for h in headers:
|
||||
v = row.get(h)
|
||||
value = getattr(v, "value", v)
|
||||
cells.append("" if value is None else str(value))
|
||||
lines.append("| " + " | ".join(cells) + " |")
|
||||
return lines
|
||||
|
||||
|
||||
def _format_chapter_data(structured_source: object | None, sheet_types: list[SheetType]) -> str:
|
||||
"""按章节定向格式化要件定义数据（design.md §6.8 ① selector=该章数据）。
|
||||
|
||||
- 命中 sheet_types 的表全部注入（章节主题数据）
|
||||
- GENERIC（自由記述）作为通用背景始终注入
|
||||
- structured_source 为 None 或无任何可注入表时返回空串
|
||||
"""
|
||||
if structured_source is None:
|
||||
return ""
|
||||
tables = getattr(structured_source, "tables", None) or []
|
||||
matched = [
|
||||
t for t in tables
|
||||
if t.detected_type in sheet_types and t.detected_type != SheetType.GENERIC
|
||||
]
|
||||
generic = [t for t in tables if t.detected_type == SheetType.GENERIC]
|
||||
selected = matched + generic
|
||||
if not selected:
|
||||
return ""
|
||||
lines: list[str] = []
|
||||
for tb in selected:
|
||||
lines.extend(_render_table(tb))
|
||||
lines.append("")
|
||||
return "\n".join(lines).rstrip()
|
||||
|
||||
|
||||
def _format_impact(report: object | None, element_type: ElementType | None = None) -> str:
|
||||
"""将影响调查书格式化为 prompt 可读文本（无报告/无分析时为空串）。
|
||||
|
||||
element_type 非 None 时仅保留该类型要素（章节级定向，design.md §6.8 ①）；
|
||||
警告始终保留（不依赖要素类型）。
|
||||
"""
|
||||
if report is None:
|
||||
return ""
|
||||
ca = getattr(report, "change_analysis", None)
|
||||
if ca is None:
|
||||
return ""
|
||||
|
||||
def keep(el) -> bool:
|
||||
return element_type is None or el.element_type == element_type.value
|
||||
|
||||
summary = getattr(report, "summary", {}) or {}
|
||||
lines = [f"project_type={getattr(ca, 'project_type', '')}"]
|
||||
lines.append(
|
||||
@@ -113,13 +199,16 @@ def _format_impact(report: object | None) -> str:
|
||||
)
|
||||
)
|
||||
for el in getattr(ca, "new_elements", []) or []:
|
||||
lines.append(f"[新規] {el.element_id} {el.element_type} {el.name}")
|
||||
if keep(el):
|
||||
lines.append(f"[新規] {el.element_id} {el.element_type} {el.name}")
|
||||
for el in getattr(ca, "modified_elements", []) or []:
|
||||
impacted = ", ".join(el.impacted_existing) or "-"
|
||||
lines.append(f"[変更] {el.element_id} {el.element_type} {el.name} → 受影响: {impacted}")
|
||||
if keep(el):
|
||||
impacted = ", ".join(el.impacted_existing) or "-"
|
||||
lines.append(f"[変更] {el.element_id} {el.element_type} {el.name} → 受影响: {impacted}")
|
||||
for el in getattr(ca, "deleted_elements", []) or []:
|
||||
impacted = ", ".join(el.impacted_existing) or "-"
|
||||
lines.append(f"[削除] {el.element_id} {el.element_type} {el.name} → 受影响: {impacted}")
|
||||
if keep(el):
|
||||
impacted = ", ".join(el.impacted_existing) or "-"
|
||||
lines.append(f"[削除] {el.element_id} {el.element_type} {el.name} → 受影响: {impacted}")
|
||||
for w in getattr(ca, "warnings", []) or []:
|
||||
lines.append(f"[警告] {w.element_id}: {w.issue}")
|
||||
return "\n".join(lines)
|
||||
|
||||
@@ -22,8 +22,12 @@ WRITER_PROMPT_TEMPLATE = (
|
||||
"设计规则:\n{{design_rules}}\n"
|
||||
"模板样式:\n{{template_styles}}\n"
|
||||
"影响调查上下文:\n{{impact}}\n"
|
||||
"参考资料:\n{{source}}\n"
|
||||
"参考资料（本章对应数据）:\n{{data}}\n"
|
||||
"请输出符合 schema 的章节内容 JSON。\n"
|
||||
"【主题约束】本章必须且仅围绕标题「{{title}}」所对应的主题撰写，"
|
||||
"严格以「参考资料（本章对应数据）」中的要件定义数据和「影响调查上下文」为核心依据；"
|
||||
"禁止输出与本章无关的系统整体架构、通用设计说明等内容，禁止套用其他章节的主题。"
|
||||
"若本章数据为空，则基于规则与影响调查上下文简要撰写，不得虚构数据。\n"
|
||||
"【语言约束】章节正文（所有 block 的 text 字段）所使用的自然语言，"
|
||||
"必须与章节标题「{{title}}」所用语言保持一致：标题为日文则用日文撰写，"
|
||||
"为中文则用中文撰写，依此类推。"
|
||||
@@ -77,16 +81,24 @@ class WriterAgent:
|
||||
return [{"index": i, "text": src[i * 1800:(i + 1) * 1800]} for i in range(n)]
|
||||
|
||||
def _resolve_prompt(self) -> Prompt:
|
||||
"""取用/注册 writer.chapter 模板。
|
||||
"""取用/注册 writer.chapter 模板，并保证返回 Prompt 对象。
|
||||
|
||||
优先使用 get_or_create（与测试 Fake 兼容）；真实 PromptRegistry 无该方法时，
|
||||
回退为 register + get。
|
||||
回退为 register + get。注意：真实 get() 返回模板字符串而非 Prompt，
|
||||
而引擎 _render_prompt 仅对 Prompt 对象做变量渲染、str 原样发送
|
||||
（否则 {{占位符}} 不被替换直接进 LLM）→ 此处统一包装为 Prompt。
|
||||
"""
|
||||
get_or_create = getattr(self.prompt_registry, "get_or_create", None)
|
||||
if get_or_create is not None:
|
||||
return get_or_create("writer.chapter", WRITER_PROMPT_TEMPLATE)
|
||||
self.prompt_registry.register("writer.chapter", "1", WRITER_PROMPT_TEMPLATE)
|
||||
return self.prompt_registry.get("writer.chapter", "1")
|
||||
resolved = get_or_create("writer.chapter", WRITER_PROMPT_TEMPLATE)
|
||||
else:
|
||||
self.prompt_registry.register("writer.chapter", "1", WRITER_PROMPT_TEMPLATE)
|
||||
resolved = self.prompt_registry.get("writer.chapter", "1")
|
||||
if isinstance(resolved, Prompt):
|
||||
return resolved
|
||||
template = getattr(resolved, "template", None) or str(resolved)
|
||||
version = str(getattr(resolved, "version", "1") or "1")
|
||||
return Prompt(name="writer.chapter", version=version, template=template)
|
||||
|
||||
def _call_llm(self, context: GenerationContext) -> dict:
|
||||
prompt = self._resolve_prompt()
|
||||
|
||||
Reference in New Issue
Block a user