Persist stage context snapshots for run workflows

docs/run-001/05-execution/final.html

@@ -201,8 +201,8 @@
         <div style="width:100%; height:100%; box-sizing:border-box; font-family:'Segoe UI',sans-serif; display:flex; flex-direction:column; gap:12px;">
   <div class="comparison-summary-card" style="background:#ffffff; border:1px solid #cbd5e1; border-radius:14px; overflow:hidden; box-shadow:0 10px 22px rgba(15,23,42,0.08);">
     <div style="padding:12px 14px; background:linear-gradient(135deg,#eff6ff 0%,#dbeafe 100%); border-bottom:1px solid #bfdbfe;">
-      <div style="font-size:11px; font-weight:700; color:#1d4ed8; letter-spacing:0.08em; text-transform:uppercase; margin-bottom:4px;">Comparison</div>
+      <div style="font-size:10.0px; font-weight:700; color:#1d4ed8; letter-spacing:0.08em; text-transform:uppercase; margin-bottom:4px;">Comparison</div>
-      <div style="font-size:16px; font-weight:800; color:#0f172a;">DX와 BIM 핵심 비교</div>
+      <div style="font-size:10.0px; font-weight:800; color:#0f172a;">DX와 BIM 핵심 비교</div>
     </div>
     <div style="padding:10px 14px 12px; display:grid; grid-template-columns:80px 1fr; row-gap:8px; column-gap:10px; font-size:10px; line-height:1.45; color:#334155;">
       <div style="font-weight:800; color:#0f172a;">범위</div><div>DX는 BIM을 포함하는 상위 개념, BIM은 3D 중심 기술</div>
@@ -213,8 +213,8 @@
   </div>
 
   <div style="background:#f8fafc; border:1px solid #cbd5e1; border-radius:14px; padding:14px; box-sizing:border-box;">
-    <div style="font-size:11px; font-weight:700; color:#475569; letter-spacing:0.08em; text-transform:uppercase; margin-bottom:6px;">Evidence</div>
+    <div style="font-size:10.0px; font-weight:700; color:#475569; letter-spacing:0.08em; text-transform:uppercase; margin-bottom:6px;">Evidence</div>
-    <div style="font-size:14px; font-weight:800; color:#0f172a; margin-bottom:8px;">정책 문서에서도 혼용</div>
+    <div style="font-size:10.0px; font-weight:800; color:#0f172a; margin-bottom:8px;">정책 문서에서도 혼용</div>
     <div style="font-size:10px; line-height:1.55; color:#475569;">• 스마트 건설 활성화 방안: 디지털화 방향 아래 BIM 전면 도입 제시</div>
     <div style="font-size:10px; line-height:1.55; color:#475569;">• 제7차 건설기술진흥 기본계획: DX 추진 방향 아래 BIM 도입 실행 과제 제시</div>
   </div>

docs/run-001/05-execution/stage_2_verification.json

@@ -0,0 +1,36 @@
+{
+  "body_bg": {
+    "passed": false,
+    "score": 0.7666666666666666,
+    "errors": [
+      "누락 문장 (7/10):",
+      "  - \"* 이로인해 BIM기술의 도입을 DX의 완성으로 오인하거나, DX를 BIM 기술 도입 수준으로 한정하는 인식...\"",
+      "  - \"* **[스마트 건설 활성화 방안(2022.07)]**\"",
+      "  - \"* 추진과제 : 건설산업 디지털화\"",
+      "  - \"* 실행과제 : BIM 전면 도입, BIM 전문인력 양성\"",
+      "  - \"* **[제7차 건설기술진흥 기본계획(2023.12)]**\""
+    ]
+  },
+  "body_core": {
+    "passed": false,
+    "score": 0.8854166666666666,
+    "errors": [
+      "누락 문장 (11/32):",
+      "  - \"| **BIM << DX**(Engineering + Management 통합) | **범위** ...\"",
+      "  - \"| **제작 및 운영**(상용 + 전용 40~80개)[Rhino, Sketchup, Blender..] + ...\"",
+      "  - \"| **공학 정보 및 콘텐츠 연계에 집중****도면, 수량, 시공계획 등 일식** | **성과품** | **...\"",
+      "  - \"| **설계/시공 생산성 혁신**(개념의 재정립) | **활용** | **3D 모델에 의한 일반적 이해 향상...\"",
+      "  - \"| **전 생애주기 활용 시스템** | **확장성** | **(설계/시공/운영) 분야별 단절** |\""
+    ]
+  },
+  "sidebar": {
+    "passed": true,
+    "score": 1.0,
+    "errors": []
+  },
+  "footer": {
+    "passed": true,
+    "score": 1.0,
+    "errors": []
+  }
+}

scripts/run_from_artifacts.py

@@ -5,7 +5,8 @@ import asyncio
 import json
 import sys
 from pathlib import Path
-DESIGN_AGENT_ROOT = Path(r'D:\\ad-hoc\\kei\\design_agent')
+
+DESIGN_AGENT_ROOT = Path(r'D:\ad-hoc\kei\design_agent')
 if str(DESIGN_AGENT_ROOT) not in sys.path:
     sys.path.insert(0, str(DESIGN_AGENT_ROOT))
 
@@ -30,6 +31,7 @@ from src.pipeline_context import (
 )
 from src.renderer import render_slide_from_html
 from src.slide_measurer import capture_slide_screenshot, measure_rendered_heights
+
 if not hasattr(html_generator, 'SIDEBAR_PROMPT') and hasattr(html_generator, '_LEGACY_SIDEBAR_PROMPT'):
     html_generator.SIDEBAR_PROMPT = html_generator._LEGACY_SIDEBAR_PROMPT
 if not hasattr(html_generator, 'FOOTER_PROMPT') and hasattr(html_generator, '_LEGACY_FOOTER_PROMPT'):
@@ -48,10 +50,12 @@ def _load_json(path: Path) -> dict:
     return json.loads(path.read_text(encoding='utf-8-sig'))
 
 
-def _build_context(content: str, base_path: str, stage1a: dict, stage1b: dict) -> PipelineContext:
+def _write_json(path: Path, data: dict) -> None:
-    ctx = create_context(content, base_path)
+    path.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding='utf-8')
 
-    normalized = normalize_mdx_content(content)
+
+def _stage_0(ctx: PipelineContext) -> PipelineContext:
+    normalized = normalize_mdx_content(ctx.raw_content)
     ctx.normalized = NormalizedContent(
         clean_text=normalized['clean_text'],
         title=normalized['title'],
@@ -60,7 +64,11 @@ def _build_context(content: str, base_path: str, stage1a: dict, stage1b: dict) -
         tables=normalized['tables'],
         sections=normalized['sections'],
     )
+    ctx.save_snapshot('stage_0')
+    return ctx
 
+
+def _stage_1a(ctx: PipelineContext, stage1a: dict) -> PipelineContext:
     analysis_raw = stage1a['analysis']
     ctx.analysis = Analysis(
         core_message=analysis_raw['core_message'],
@@ -68,15 +76,21 @@ def _build_context(content: str, base_path: str, stage1a: dict, stage1b: dict) -
         total_pages=analysis_raw.get('total_pages', 1),
     )
     ctx.page_structure = PageStructure(roles=stage1a['page_structure'])
+    ctx.topics = [Topic(**raw) for raw in stage1a['topics']]
+    ctx.save_snapshot('stage_1a')
+    return ctx
 
+
+def _stage_1b(ctx: PipelineContext, stage1b: dict) -> PipelineContext:
     refined_map = {item['topic_id']: item for item in stage1b['concepts']}
     topics = []
-    for raw in stage1a['topics']:
+    for raw in ctx.topics:
-        merged = dict(raw)
+        merged = raw.model_dump()
-        if raw['id'] in refined_map:
+        if raw.id in refined_map:
-            merged.update(refined_map[raw['id']])
+            merged.update(refined_map[raw.id])
         topics.append(Topic(**merged))
     ctx.topics = topics
+    ctx.save_snapshot('stage_1b')
     return ctx
 
 
@@ -138,6 +152,7 @@ def _stage_1_5a(ctx: PipelineContext) -> PipelineContext:
         })
     ctx.slide_images = slide_images
     ctx.analysis = ctx.analysis.model_copy(update={'image_sizes': image_sizes or {}})
+    ctx.save_snapshot('stage_1_5a')
     return ctx
 
 
@@ -157,6 +172,7 @@ def _stage_1_7(ctx: PipelineContext) -> PipelineContext:
         )
         for role, ref in refs_raw.items()
     }
+    ctx.save_snapshot('stage_1_7')
     return ctx
 
 
@@ -184,6 +200,7 @@ def _stage_1_5b(ctx: PipelineContext) -> PipelineContext:
             )
         })
     ctx.containers = updated
+    ctx.save_snapshot('stage_1_5b')
     return ctx
 
 
@@ -218,7 +235,7 @@ async def _stage_2(ctx: PipelineContext) -> PipelineContext:
             for role, ci in ctx.containers.items()
         },
     }
-    generated, _verification = await generate_with_retry(
+    generated, verification = await generate_with_retry(
         content=ctx.raw_content,
         analysis=analysis_dict,
         container_specs=container_specs_dict,
@@ -226,6 +243,16 @@ async def _stage_2(ctx: PipelineContext) -> PipelineContext:
         images=ctx.slide_images,
     )
     ctx.generated_html = generated
+    verification_path = ctx.get_run_dir() / 'stage_2_verification.json'
+    _write_json(verification_path, {
+        area: {
+            'passed': result.passed,
+            'score': result.score,
+            'errors': result.errors,
+        }
+        for area, result in verification.items()
+    })
+    ctx.save_snapshot('stage_2')
     return ctx
 
 
@@ -239,15 +266,17 @@ def _stage_3(ctx: PipelineContext) -> PipelineContext:
     ctx.rendered_html = render_slide_from_html(ctx.generated_html, analysis_dict, ctx.preset)
     if ctx.base_path:
         ctx.rendered_html = embed_images(ctx.rendered_html, ctx.base_path)
+    ctx.save_snapshot('stage_3')
     return ctx
 
 
-def _stage_4_lite(ctx: PipelineContext) -> PipelineContext:
+def _stage_4(ctx: PipelineContext) -> PipelineContext:
     ctx.measurement = measure_rendered_heights(ctx.rendered_html)
     ctx.screenshot_b64 = capture_slide_screenshot(ctx.rendered_html) or ''
     ctx.quality_score = 100 if not any(
         zone.get('overflowed') for zone in ctx.measurement.get('zones', {}).values()
     ) else 60
+    ctx.save_snapshot('stage_4')
     return ctx
 
 
@@ -264,16 +293,22 @@ async def main() -> None:
     stage1a = _load_json(Path(args.stage1a))
     stage1b = _load_json(Path(args.stage1b))
 
-    ctx = _build_context(content, args.base_path, stage1a, stage1b)
+    out_dir = Path(args.output_dir)
+    out_dir.mkdir(parents=True, exist_ok=True)
+
+    ctx = create_context(content, args.base_path)
+    ctx.run_dir = str(out_dir)
+
+    ctx = _stage_0(ctx)
+    ctx = _stage_1a(ctx, stage1a)
+    ctx = _stage_1b(ctx, stage1b)
     ctx = _stage_1_5a(ctx)
     ctx = _stage_1_7(ctx)
     ctx = _stage_1_5b(ctx)
     ctx = await _stage_2(ctx)
     ctx = _stage_3(ctx)
-    ctx = _stage_4_lite(ctx)
+    ctx = _stage_4(ctx)
 
-    out_dir = Path(args.output_dir)
-    out_dir.mkdir(parents=True, exist_ok=True)
     (out_dir / 'generated_html.json').write_text(
         json.dumps(ctx.generated_html, ensure_ascii=False, indent=2),
         encoding='utf-8',
@@ -287,10 +322,11 @@ async def main() -> None:
         ctx.model_dump_json(indent=2, exclude={'screenshot_b64', 'rendered_html'}),
         encoding='utf-8',
     )
+    (out_dir / 'final_context.json').write_text(
+        ctx.model_dump_json(indent=2, exclude={'screenshot_b64', 'rendered_html'}),
+        encoding='utf-8',
+    )
 
 
 if __name__ == '__main__':
     asyncio.run(main())
-
-
-