release: v0.20 Codex-ready skill-driven core

tests/test_chapter_assembly.py

@@ -62,6 +62,10 @@ def write_packet(path: Path, task_id: str, claim: str, source_id: str) -> None:
         "evidence_items": [{"source_id": source_id, "summary": f"{claim} 的证据"}],
         "counter_evidence": [{"claim": "仍需关注样本量和外推限制", "source_ids": ["src_counter"]}],
         "source_ids": [source_id, "src_counter"],
+        "sources": [
+            {"id": source_id, "title": "来源", "url": f"https://example.com/{source_id}"},
+            {"id": "src_counter", "title": "反方来源", "url": "https://example.com/counter"},
+        ],
         "source_quality_notes": [f"{source_id} Tier 1"],
         "open_questions": ["还需要补充中国市场数据"],
         "raw_quotes_or_notes": ["English note can remain as source material."],
@@ -76,6 +80,7 @@ def test_build_chapter_briefs_aggregates_packets_by_chapter(tmp_path: Path) -> N
         {
             "task_id": "ch01-clinical",
             "chapter_ids": ["ch01"],
+            "chapter_title": "临床证据正在重塑需求判断",
             "topic_axis": "clinical",
             "questions": ["q"],
             "search_routes": ["scholar"],
@@ -84,6 +89,7 @@ def test_build_chapter_briefs_aggregates_packets_by_chapter(tmp_path: Path) -> N
         {
             "task_id": "ch01-market",
             "chapter_ids": ["ch01"],
+            "chapter_title": "临床证据正在重塑需求判断",
             "topic_axis": "market",
             "questions": ["q"],
             "search_routes": ["news"],
@@ -101,6 +107,7 @@ def test_build_chapter_briefs_aggregates_packets_by_chapter(tmp_path: Path) -> N
     brief = briefs[0]
     validate_chapter_brief(brief)
     assert brief["chapter_id"] == "ch01"
+    assert brief["chapter_title"] == "临床证据正在重塑需求判断"
     assert brief["packet_ids"] == ["ch01-clinical", "ch01-market"]
     assert "src_001" in brief["source_ids"]
     assert "src_002" in brief["source_ids"]
@@ -118,6 +125,50 @@ def test_build_chapter_briefs_aggregates_packets_by_chapter(tmp_path: Path) -> N
     assert (project / "phase2/compressed_findings/ch01.json").exists()
 
 
+def test_build_chapter_briefs_skips_placeholder_packets(tmp_path: Path) -> None:
+    project = tmp_path / "project"
+    cards = [
+        {
+            "task_id": "ch01-good",
+            "chapter_ids": ["ch01"],
+            "chapter_title": "临床证据正在重塑需求判断",
+            "topic_axis": "clinical",
+            "questions": ["q"],
+            "search_routes": ["scholar"],
+            "output_packet": "phase2/packets/ch01-good.json",
+        },
+        {
+            "task_id": "ch01-empty",
+            "chapter_ids": ["ch01"],
+            "chapter_title": "临床证据正在重塑需求判断",
+            "topic_axis": "clinical",
+            "questions": ["q"],
+            "search_routes": ["scholar"],
+            "output_packet": "phase2/packets/ch01-empty.json",
+        },
+    ]
+    (project / "phase2").mkdir(parents=True)
+    (project / "phase2/task_cards.json").write_text(json.dumps(cards, ensure_ascii=False), encoding="utf-8")
+    write_packet(project / "phase2/packets/ch01-good.json", "ch01-good", "临床证据支持核心判断", "src_001")
+    empty = {
+        "task_id": "ch01-empty",
+        "claims": [],
+        "evidence_items": [],
+        "counter_evidence": [],
+        "source_ids": [],
+        "source_quality_notes": [],
+        "open_questions": [],
+        "raw_quotes_or_notes": [],
+    }
+    (project / "phase2/packets/ch01-empty.json").write_text(json.dumps(empty, ensure_ascii=False), encoding="utf-8")
+
+    briefs = build_chapter_briefs(project)
+
+    assert len(briefs) == 1
+    warnings = json.loads((project / "phase2/brief_warnings.json").read_text(encoding="utf-8"))
+    assert warnings[0]["task_id"] == "ch01-empty"
+
+
 def test_chapter_prompt_contains_brief_and_fragmentation_guard() -> None:
     brief = {
         "chapter_id": "ch01",
@@ -161,6 +212,8 @@ def test_chapter_assembly_worker_writes_markdown(tmp_path: Path) -> None:
     assert output == tmp_path / "phase2/drafts/ch01.md"
     assert "结论先行" in output.read_text(encoding="utf-8")
     assert fake.calls[0]["model"] == role.model
+    assert "章节证据分析师" in fake.calls[0]["system"]
+    assert "中文章节组装 worker" in fake.calls[0]["system"]
 
 
 def test_validate_chapter_markdown_rejects_unknown_source_ids() -> None:

tests/test_phase0_materials.py

@@ -68,13 +68,22 @@ def test_framework_mentions_ingested_materials(tmp_path: Path) -> None:
 
     assert "phase0/extracted/audit.md" in framework
     assert "NMPA、FDA、EMA、ICH、WHO" in framework
+    assert "Phase1 的职责是大胆假设" in framework
+    assert "本章要解决的问题" in framework
     assert "请先确认 `phase1/material_brief.md`" in framework
     assert research_brief_md.exists()
     assert "任务切分原则" in research_brief_md.read_text(encoding="utf-8")
+    assert "章节命题与求证计划" in research_brief_md.read_text(encoding="utf-8")
+    assert (project / "phase1" / "hypothesis_map.json").exists()
     assert brief["research_method"] == "gmp_quality_operations_diagnosis"
     assert brief["work_language"] == "zh"
+    assert brief["phase2_mode"] == "chapter_integrated"
+    assert brief["central_thesis"]
+    assert brief["chapter_planning"][0]["phase2_prompt_context"]
     assert brief["task_planning"]["required_skills"]
-    assert brief["task_planning"]["search_routes_by_axis"]["counter"] == ["scholar", "general"]
+    assert brief["task_planning"]["phase2_mode"] == "chapter_integrated"
+    assert brief["task_planning"]["search_routes_by_axis"]["counter"] == ["fda", "scholar", "evidence", "general"]
+    assert brief["task_planning"]["search_routes_by_axis"]["quality_system_gap"] == ["fda", "evidence", "general"]
     assert brief["phase2_inputs"]["framework_path"] == "phase1/framework.md"
 
 

tests/test_phase3_model_review.py

@@ -0,0 +1,80 @@
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+if str(REPO_ROOT) not in sys.path:
+    sys.path.insert(0, str(REPO_ROOT))
+
+from scripts.runtime.review import (
+    build_phase3_model_review_context,
+    build_phase3_model_critique,
+    phase3_model_review_system_prompt,
+)
+
+
+class FakeClient:
+    def __init__(self) -> None:
+        self.calls: list[dict[str, object]] = []
+
+    def chat_complete(self, **kwargs) -> str:
+        self.calls.append(kwargs)
+        return "# Phase 3 Opus 4.7 独立审校\n\n## 总体判定\n\n回炉 Phase2。\n"
+
+
+def make_project(tmp_path: Path) -> Path:
+    project = tmp_path / "project"
+    (project / "phase1").mkdir(parents=True)
+    (project / "phase2/drafts").mkdir(parents=True)
+    (project / "phase2/compressed_findings").mkdir(parents=True)
+    (project / "manifest.json").write_text(
+        json.dumps({"topic": "白帆测试项目", "phase3": {}}, ensure_ascii=False),
+        encoding="utf-8",
+    )
+    (project / "phase1/framework.md").write_text("## 第1章 质量体系判断\n", encoding="utf-8")
+    (project / "phase2/drafts/ch01.md").write_text("## 质量体系判断\n\n正文。[src_001]\n", encoding="utf-8")
+    (project / "phase2/sources.jsonl").write_text('{"id":"src_001","title":"来源","url":"https://www.fda.gov/example"}\n', encoding="utf-8")
+    (project / "phase2/compressed_findings/ch01.json").write_text(
+        json.dumps(
+            {
+                "chapter_id": "ch01",
+                "chapter_title": "质量体系判断",
+                "packet_ids": ["ch01-a"],
+                "chapter_thesis": "质量体系需要补证据",
+                "key_findings": [],
+                "evidence_landings": [],
+                "counter_evidence": [],
+                "source_ids": ["src_001"],
+                "open_questions": [],
+                "writing_plan": [],
+            },
+            ensure_ascii=False,
+        ),
+        encoding="utf-8",
+    )
+    return project
+
+
+def test_phase3_model_context_contains_structured_inputs(tmp_path: Path) -> None:
+    project = make_project(tmp_path)
+
+    context = build_phase3_model_review_context(project)
+
+    assert "Deterministic Review Baseline" in context
+    assert "Compressed Findings" in context
+    assert "Chapter Drafts" in context
+    assert "src_001" in context
+
+
+def test_phase3_model_review_calls_requested_model_and_writes_critique(tmp_path: Path) -> None:
+    project = make_project(tmp_path)
+    fake = FakeClient()
+
+    out = build_phase3_model_critique(project, client=fake, model="zenmux-anthropic/claude-opus-4-7")
+
+    assert out.exists()
+    assert fake.calls[0]["model"] == "zenmux-anthropic/claude-opus-4-7"
+    assert "独立总编审校" in fake.calls[0]["system"]
+    assert "FDA/NMPA/EMA/ICH/WHO" in phase3_model_review_system_prompt()

tests/test_polish_prompt.py

@@ -0,0 +1,19 @@
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+if str(REPO_ROOT) not in sys.path:
+    sys.path.insert(0, str(REPO_ROOT))
+
+from scripts.polish import build_polish_system_prompt
+
+
+def test_polish_system_prompt_loads_humanizer_and_output_hygiene() -> None:
+    prompt = build_polish_system_prompt()
+
+    assert "# Skill: humanizer-cn" in prompt
+    assert "CN-1" in prompt
+    assert "# Skill: output-hygiene" in prompt
+    assert "禁止词" in prompt

tests/test_reporting.py

@@ -10,6 +10,7 @@ if str(REPO_ROOT) not in sys.path:
 
 from scripts.reporting.fonts import resolve_quarto_fonts
 from scripts.reporting.references import build_references_block
+from scripts.number_citations import number_citations
 
 
 def test_build_references_block_uses_only_cited_sources(tmp_path: Path) -> None:
@@ -32,3 +33,38 @@ def test_resolve_quarto_fonts_returns_stable_defaults_for_missing_dir(tmp_path:
     assert fonts.main_font == "Source Han Serif CN"
     assert fonts.sans_font == "Source Han Sans CN"
     assert fonts.requires_system_fonts is True
+
+
+def test_number_citations_replaces_source_ids_and_keeps_url() -> None:
+    text = "# 报告\n\n关键判断。[src_a, src_b]\n\n## 参考文献\n\n旧列表\n"
+    sources = {
+        "src_a": {"title": "法规 A", "url": "https://example.com/a"},
+        "src_b": {"title": "指南 B", "url": "https://example.com/b"},
+    }
+
+    numbered, records = number_citations(text=text, sources=sources)
+
+    assert "关键判断。<sup>[1, 2]</sup>" in numbered
+    assert "[src_a" not in numbered
+    assert "## 参考来源清单" in numbered
+    assert "1. 法规 A. https://example.com/a" in numbered
+    assert "旧列表" not in numbered
+    assert [record["source_id"] for record in records] == ["src_a", "src_b"]
+
+
+def test_number_citations_deduplicates_same_underlying_source() -> None:
+    text = "甲。[src_a]\n\n乙。[src_b, src_c]\n"
+    sources = {
+        "src_a": {"title": "同一报告 OCR", "path": "phase0/report.md"},
+        "src_b": {"title": "同一报告", "path": "phase0/report.md"},
+        "src_c": {"title": "法规 C", "url": "https://example.com/c"},
+    }
+
+    numbered, records = number_citations(text=text, sources=sources)
+
+    assert "甲。<sup>[1]</sup>" in numbered
+    assert "乙。<sup>[1, 2]</sup>" in numbered
+    assert numbered.count("同一报告") == 1
+    assert "同一报告 OCR" not in numbered
+    assert len(records) == 2
+    assert records[0]["source_ids"] == ["src_a", "src_b"]

tests/test_search_grounded_packets.py

@@ -11,7 +11,7 @@ if str(REPO_ROOT) not in sys.path:
 from scripts.runtime.roles import resolve_runtime_profile
 from scripts.runtime.sources import append_packet_sources, rebuild_sources_from_packets
 from scripts.runtime.tasks import TaskCard
-from scripts.runtime.workers import PacketWorker, build_search_context
+from scripts.runtime.workers import PacketWorker, build_material_context, build_route_query, build_search_context, normalize_packet_against_context
 
 
 class FakeSearchProvider:
@@ -57,6 +57,66 @@ def test_build_search_context_assigns_stable_source_ids() -> None:
     assert context["routes_used"] == ["scholar", "general"]
 
 
+def test_fda_route_query_uses_english_axis_terms_not_chinese_title() -> None:
+    card = TaskCard(
+        task_id="ch10-fda_enforcement_precedents",
+        chapter_ids=["ch10"],
+        topic_axis="fda_enforcement_precedents",
+        questions=["立即纠偏、体系补强、能力建设三层整改路线图必须绑定 owner、关闭证据和复核机制"],
+        search_routes=["fda"],
+        output_packet="phase2/packets/ch10-fda_enforcement_precedents.json",
+        chapter_title="立即纠偏、体系补强、能力建设三层整改路线图必须绑定 owner、关闭证据和复核机制",
+    )
+
+    query = build_route_query(card, "fda")
+
+    assert "立即纠偏" not in query
+    assert "CAPA" in query
+    assert "remediation" in query
+    assert "verification evidence" in query
+
+
+def test_integrated_scholar_query_does_not_leak_internal_axis_or_cjk_punctuation() -> None:
+    card = TaskCard(
+        task_id="ch07-chapter_integrated",
+        chapter_ids=["ch07"],
+        topic_axis="chapter_integrated",
+        questions=["人员能力：培训有效性比培训记录更关键"],
+        search_routes=["scholar"],
+        output_packet="phase2/packets/ch07-chapter_integrated.json",
+        chapter_title="人员能力：培训有效性比培训记录更关键",
+    )
+
+    query = build_route_query(card, "scholar")
+
+    assert "chapter_integrated" not in query
+    assert "、" not in query
+    assert "  " not in query
+    assert "training" in query
+    assert "quality" in query
+    assert not any("\u4e00" <= char <= "\u9fff" for char in query)
+
+
+def test_evidence_route_query_is_short_english_candidate_evidence_query() -> None:
+    card = TaskCard(
+        task_id="ch08-chapter_integrated",
+        chapter_ids=["ch08"],
+        topic_axis="chapter_integrated",
+        questions=["运营管理需要建立跨部门节奏、问题升级、指标看板和管理层 review"],
+        search_routes=["evidence"],
+        output_packet="phase2/packets/ch08-chapter_integrated.json",
+        chapter_title="运营管理需要建立跨部门节奏、问题升级、指标看板和管理层 review",
+    )
+
+    query = build_route_query(card, "evidence")
+
+    assert "evidence" in query
+    assert "quality" in query
+    assert "operations" in query
+    assert "运营管理" not in query
+    assert not any("\u4e00" <= char <= "\u9fff" for char in query)
+
+
 def test_packet_worker_includes_search_context_in_prompt() -> None:
     context = build_search_context(sample_card(), FakeSearchProvider(), num_results_per_route=1)
     response = {
@@ -79,7 +139,71 @@ def test_packet_worker_includes_search_context_in_prompt() -> None:
     assert "candidate_sources" in fake.calls[0]["user"]
 
 
-def test_append_packet_sources_dedupes_by_url(tmp_path: Path) -> None:
+def test_normalize_packet_fills_source_ids_and_sources_from_context() -> None:
+    context = {
+        "candidate_sources": [
+            {"id": "src_a", "title": "A", "url": "https://example.com/a", "tier": "Tier 2", "score": 7}
+        ]
+    }
+    packet = {
+        "task_id": "ch01",
+        "claims": [{"claim": "判断", "source_ids": ["src_a"]}],
+        "evidence_items": [{"source_id": "src_a", "summary": "证据"}],
+        "counter_evidence": [{"claim": "反方", "source_ids": ["src_a"]}],
+        "source_quality_notes": [],
+        "open_questions": [],
+        "raw_quotes_or_notes": [],
+    }
+
+    normalized = normalize_packet_against_context(packet, context, None)
+
+    assert normalized["source_ids"] == ["src_a"]
+    assert normalized["sources"] == context["candidate_sources"]
+
+
+def test_material_context_is_loaded_and_allowed_as_source(tmp_path: Path) -> None:
+    project = tmp_path / "project"
+    material = project / "phase0/extracted/audit.md"
+    material.parent.mkdir(parents=True)
+    material.write_text("白帆现场发现：偏差调查未闭环。", encoding="utf-8")
+    card = TaskCard(
+        task_id="ch01-chapter_integrated",
+        chapter_ids=["ch01"],
+        topic_axis="chapter_integrated",
+        questions=["q"],
+        search_routes=[],
+        output_packet="phase2/packets/ch01-chapter_integrated.json",
+        allowed_materials=["phase0/extracted/audit.md"],
+    )
+    context = build_material_context(card, project)
+    response = {
+        "task_id": "ch01-chapter_integrated",
+        "claims": [{"claim": "现场材料显示偏差调查需要补强", "source_ids": [context["materials"][0]["source_id"]]}],
+        "evidence_items": [{"source_id": context["materials"][0]["source_id"], "summary": "偏差调查未闭环。"}],
+        "counter_evidence": [{"claim": "需与完整审计报告交叉确认", "source_ids": [context["materials"][0]["source_id"]]}],
+        "source_ids": [context["materials"][0]["source_id"]],
+        "sources": [
+            {
+                "id": context["materials"][0]["source_id"],
+                "title": "audit.md",
+                "url": "phase0/extracted/audit.md",
+                "tier": "local_material",
+            }
+        ],
+        "source_quality_notes": ["本地材料作为起点证据"],
+        "open_questions": [],
+        "raw_quotes_or_notes": ["白帆现场发现：偏差调查未闭环。"],
+    }
+    fake = FakeClient(response)
+    role = resolve_runtime_profile(profile="medium").role_for_task("evidence_packet")
+
+    packet = PacketWorker(role=role, client=fake, project_root=project).run(card)
+
+    assert packet["source_ids"] == [context["materials"][0]["source_id"]]
+    assert "白帆现场发现" in fake.calls[0]["user"]
+
+
+def test_append_packet_sources_preserves_distinct_source_ids_for_same_url(tmp_path: Path) -> None:
     packet = {
         "sources": [
             {"id": "src_a", "title": "A", "url": "https://example.com/a", "tier": "Tier 2", "score": 7},
@@ -89,11 +213,11 @@ def test_append_packet_sources_dedupes_by_url(tmp_path: Path) -> None:
 
     written = append_packet_sources(tmp_path / "sources.jsonl", packet)
 
-    assert written == 1
-    assert len((tmp_path / "sources.jsonl").read_text(encoding="utf-8").splitlines()) == 1
+    assert written == 2
+    assert len((tmp_path / "sources.jsonl").read_text(encoding="utf-8").splitlines()) == 2
 
 
-def test_rebuild_sources_from_packets_dedupes_manual_packets(tmp_path: Path) -> None:
+def test_rebuild_sources_from_packets_preserves_distinct_source_ids(tmp_path: Path) -> None:
     project = tmp_path / "project"
     packets = project / "phase2" / "packets"
     packets.mkdir(parents=True)
@@ -109,7 +233,34 @@ def test_rebuild_sources_from_packets_dedupes_manual_packets(tmp_path: Path) ->
     count = rebuild_sources_from_packets(project)
 
     lines = (project / "phase2" / "sources.jsonl").read_text(encoding="utf-8").splitlines()
-    assert count == 2
-    assert len(lines) == 2
+    assert count == 3
+    assert len(lines) == 3
     assert "src_001" in lines[0]
-    assert "src_003" in lines[1]
+    assert "src_002" in lines[1]
+    assert "src_003" in lines[2]
+
+
+def test_rebuild_sources_preserves_cache_metadata(tmp_path: Path) -> None:
+    project = tmp_path / "project"
+    packets = project / "phase2" / "packets"
+    packets.mkdir(parents=True)
+    source = {"id": "src_001", "title": "A", "url": "https://example.com/a"}
+    (packets / "ch01-a.json").write_text(json.dumps({"sources": [source]}, ensure_ascii=False), encoding="utf-8")
+    (project / "phase2/sources.jsonl").write_text(
+        json.dumps(
+            {
+                **source,
+                "cached_text_path": "phase2/source_cache/md/src_001.md",
+                "cache_status": "fetched",
+            },
+            ensure_ascii=False,
+        )
+        + "\n",
+        encoding="utf-8",
+    )
+
+    rebuild_sources_from_packets(project)
+
+    row = json.loads((project / "phase2/sources.jsonl").read_text(encoding="utf-8"))
+    assert row["cached_text_path"] == "phase2/source_cache/md/src_001.md"
+    assert row["cache_status"] == "fetched"

tests/test_source_cache.py

@@ -0,0 +1,70 @@
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+if str(REPO_ROOT) not in sys.path:
+    sys.path.insert(0, str(REPO_ROOT))
+
+from scripts.runtime.source_cache import cache_sources, is_important_source
+
+
+class FakeResponse:
+    headers = {"content-type": "text/html; charset=utf-8"}
+    url = "https://www.fda.gov/example"
+    content = b"<html><body><h1>FDA Guidance</h1><p>Important CGMP text.</p></body></html>"
+
+    def raise_for_status(self) -> None:
+        return None
+
+
+class FakeClient:
+    def __enter__(self) -> "FakeClient":
+        return self
+
+    def __exit__(self, *_args) -> None:
+        return None
+
+    def get(self, url: str) -> FakeResponse:
+        assert url == "https://www.fda.gov/example"
+        return FakeResponse()
+
+    def close(self) -> None:
+        return None
+
+
+def test_is_important_source_detects_official_regulator() -> None:
+    assert is_important_source({"url": "https://www.fda.gov/example", "title": "FDA"})
+    assert not is_important_source({"url": "https://example.com/blog", "title": "Blog"})
+
+
+def test_cache_sources_writes_markdown_and_updates_registry(tmp_path: Path, monkeypatch) -> None:
+    project = tmp_path / "project"
+    sources = project / "phase2" / "sources.jsonl"
+    sources.parent.mkdir(parents=True)
+    sources.write_text(
+        json.dumps(
+            {
+                "id": "src_fda_001",
+                "title": "FDA Guidance",
+                "url": "https://www.fda.gov/example",
+                "tier": "Tier 1",
+            },
+            ensure_ascii=False,
+        )
+        + "\n",
+        encoding="utf-8",
+    )
+
+    monkeypatch.setattr("scripts.runtime.source_cache.httpx.Client", lambda **_kwargs: FakeClient())
+
+    results = cache_sources(project)
+
+    rows = [json.loads(line) for line in sources.read_text(encoding="utf-8").splitlines()]
+    assert len(results) == 1
+    assert rows[0]["cached_text_path"].startswith("phase2/source_cache/md/")
+    cached = project / rows[0]["cached_text_path"]
+    assert cached.exists()
+    assert "Important CGMP text." in cached.read_text(encoding="utf-8")

tests/test_v020_cli.py

@@ -68,6 +68,10 @@ def test_research_build_briefs_does_not_overwrite_existing_packets(tmp_path: Pat
         "evidence_items": [{"source_id": "src_001", "summary": "证据"}],
         "counter_evidence": [{"claim": "限制", "source_ids": ["src_002"]}],
         "source_ids": ["src_001", "src_002"],
+        "sources": [
+            {"id": "src_001", "title": "来源1", "url": "https://example.com/1"},
+            {"id": "src_002", "title": "来源2", "url": "https://example.com/2"},
+        ],
         "source_quality_notes": ["src_001 Tier 1"],
         "open_questions": [],
         "raw_quotes_or_notes": [],
@@ -80,6 +84,28 @@ def test_research_build_briefs_does_not_overwrite_existing_packets(tmp_path: Pat
     assert "真实证据不能被 skeleton 覆盖" in packet_path.read_text(encoding="utf-8")
 
 
+def test_codex_native_profile_does_not_claim_python_core_model_execution(tmp_path: Path) -> None:
+    project = tmp_path / "project"
+    (project / "phase1").mkdir(parents=True)
+    (project / "manifest.json").write_text(
+        '{"research_method": "mckinsey_market", "phase1": {"approved": true}, "phase2": {}}\n',
+        encoding="utf-8",
+    )
+    (project / "phase1/framework.md").write_text(
+        "## 第1章 临床证据正在重塑需求判断\n\n研究思路。",
+        encoding="utf-8",
+    )
+
+    args = dr.build_parser().parse_args(["research", str(project), "--profile", "codex_native", "--execute-packets"])
+
+    try:
+        dr.cmd_research(args)
+    except SystemExit as exc:
+        assert "not Codex App built-in models" in str(exc)
+    else:
+        raise AssertionError("codex_native must not execute through Python external clients")
+
+
 def test_packet_state_counts_ignores_stale_errors_for_ready_packets(tmp_path: Path) -> None:
     project = tmp_path / "project"
     (project / "phase2/packets").mkdir(parents=True)
@@ -184,6 +210,30 @@ def test_init_and_frame_create_executable_python_core_project(tmp_path: Path) ->
     assert "中文" in framework
 
 
+def test_frame_can_preserve_existing_outline(tmp_path: Path) -> None:
+    project = tmp_path / "custom-outline"
+    (project / "phase1").mkdir(parents=True)
+    (project / "manifest.json").write_text(
+        '{"topic": "自定义研究", "research_method": "mckinsey_market", "target_words": 12000, "phase1": {}}\n',
+        encoding="utf-8",
+    )
+    (project / "phase1/framework.md").write_text(
+        "## 第1章 第一条自定义主线\n\n## 第2章 第二条自定义主线\n\n## 第3章 第三条自定义主线\n\n"
+        "## 第4章 第四条自定义主线\n\n## 第5章 第五条自定义主线\n\n## 第6章 第六条自定义主线\n\n"
+        "## 第7章 第七条自定义主线\n\n## 第8章 第八条自定义主线\n",
+        encoding="utf-8",
+    )
+
+    args = dr.build_parser().parse_args(["frame", str(project), "--preserve-existing-outline"])
+
+    assert dr.cmd_frame(args) == 0
+    framework = (project / "phase1/framework.md").read_text(encoding="utf-8")
+    brief = json.loads((project / "phase1/research_brief.json").read_text(encoding="utf-8"))
+    assert "第一条自定义主线" in framework
+    assert "本章要解决的问题" in framework
+    assert brief["chapter_planning"][0]["title"] == "第一条自定义主线"
+
+
 def test_review_writes_phase3_critique(tmp_path: Path) -> None:
     project = tmp_path / "project"
     (project / "phase1").mkdir(parents=True)
@@ -205,6 +255,54 @@ def test_review_writes_phase3_critique(tmp_path: Path) -> None:
     assert "src_001" in text
 
 
+def test_review_model_dry_run_exposes_opus_context_plan(tmp_path: Path, capsys) -> None:
+    project = tmp_path / "project"
+    project.mkdir()
+    (project / "manifest.json").write_text('{"topic": "测试项目"}\n', encoding="utf-8")
+
+    args = dr.build_parser().parse_args(["review", str(project), "--model-review", "--dry-run"])
+
+    assert dr.cmd_review(args) == 0
+    out = capsys.readouterr().out
+    assert "zenmux-anthropic/claude-opus-4-7" in out
+    assert "review_context_opus_4_7.md" in out
+
+
+def test_finalize_polish_dry_run_uses_polish_source_argument(tmp_path: Path, capsys) -> None:
+    project = tmp_path / "project"
+    (project / "phase4").mkdir(parents=True)
+    (project / "manifest.json").write_text(
+        '{"model_profile": "medium", "phase4": {}}\n',
+        encoding="utf-8",
+    )
+    (project / "phase4/final_zh.md").write_text("# 中文终稿\n", encoding="utf-8")
+
+    args = dr.build_parser().parse_args(["finalize", str(project), "--polish", "--dry-run"])
+
+    assert dr.cmd_finalize(args) == 0
+    out = capsys.readouterr().out
+    assert "scripts/polish.py" in out
+    assert "--source phase4/final_zh.md" in out
+    assert "--input phase4/final_zh.md" not in out.split("scripts/polish.py", 1)[1]
+
+
+def test_finalize_number_citations_dry_run_builds_numbered_markdown(tmp_path: Path, capsys) -> None:
+    project = tmp_path / "project"
+    (project / "phase4").mkdir(parents=True)
+    (project / "manifest.json").write_text(
+        '{"model_profile": "medium", "phase4": {}}\n',
+        encoding="utf-8",
+    )
+    (project / "phase4/final_zh.md").write_text("# 中文终稿\n\n正文。[src_001]\n", encoding="utf-8")
+
+    args = dr.build_parser().parse_args(["finalize", str(project), "--number-citations", "--dry-run"])
+
+    assert dr.cmd_finalize(args) == 0
+    out = capsys.readouterr().out
+    assert "scripts/number_citations.py" in out
+    assert "--input phase4/final_zh_numbered.md" in out
+
+
 def test_run_new_topic_initializes_and_frames_project(tmp_path: Path) -> None:
     args = dr.build_parser().parse_args(
         [

tests/test_v020_runtime.py

@@ -12,6 +12,8 @@ if str(REPO_ROOT) not in sys.path:
 
 from scripts.lib.model_config import resolve_model_profile
 from scripts.runtime.roles import resolve_runtime_profile
+from scripts.runtime.methods import ResearchMethodRegistry
+from scripts.runtime.phase1 import build_chapter_planning
 from scripts.runtime.skills import SkillRegistry
 from scripts.runtime.tasks import (
     TaskCard,
@@ -78,6 +80,7 @@ def test_generate_task_cards_from_chinese_framework() -> None:
         "ch02-regulatory",
     ]
     assert cards[0].output_packet == "phase2/packets/ch01-clinical.json"
+    assert cards[0].chapter_title == "GLP-1 产业链的增量来自适应症扩张"
     assert cards[0].research_goal
     assert "search-gateway" in cards[0].required_skills
     assert cards[0].expected_evidence["min_tier_1_2_sources"] == 2
@@ -121,6 +124,122 @@ def test_generate_task_cards_from_research_brief_carries_prompt_and_skills() ->
     assert "search-gateway" in cards[0].required_skills
 
 
+def test_gmp_task_cards_include_fda_enforcement_route() -> None:
+    brief = {
+        "research_method": "gmp_quality_operations_diagnosis",
+        "task_planning": {
+            "search_routes_by_axis": {
+                "quality_system_gap": ["fda", "general"],
+            },
+        },
+    }
+    framework = "## 第1章 偏差和 CAPA 闭环能力决定质量体系可信度\n\n研究思路。"
+
+    cards = generate_task_cards_from_research_brief(
+        "baifan-test",
+        framework,
+        brief,
+        axes=["quality_system_gap"],
+    )
+
+    assert cards[0].search_routes == ["fda", "general"]
+    assert "FDA Warning Letters" in " ".join(cards[0].questions)
+    assert "fda_warning_letter_or_meeting_record" in cards[0].expected_evidence["preferred_evidence_types"]
+
+
+def test_integrated_chapter_mode_is_method_driven_not_gmp_hardcoded() -> None:
+    brief = {
+        "research_method": "mckinsey_market",
+        "phase2_mode": "chapter_integrated",
+        "task_planning": {},
+    }
+    framework = "## 第1章 市场需求正在被支付政策重塑\n\n研究思路。"
+
+    cards = generate_task_cards_from_research_brief(
+        "market-test",
+        framework,
+        brief,
+    )
+
+    assert [card.task_id for card in cards] == ["ch01-chapter_integrated"]
+    assert "literature evidence" in " ".join(cards[0].questions)
+    assert "FDA Warning Letters" not in " ".join(cards[0].questions)
+
+
+def test_integrated_task_card_uses_phase1_chapter_planning() -> None:
+    brief = {
+        "research_method": "gmp_quality_operations_diagnosis",
+        "phase2_mode": "chapter_integrated",
+        "chapter_planning": [
+            {
+                "chapter_id": "ch01",
+                "title": "审计发现应先转化为商业化阶段门缺口",
+                "core_question": "本章要判断审计发现是否反映阶段门缺口。",
+                "bold_hypothesis": "大胆假设：风险项计数低估了商业化 readiness 缺口。",
+                "verification_plan": ["提取现场材料原文", "检索官方法规和执法案例"],
+                "evidence_lanes": ["site audit findings", "official baseline"],
+                "minimum_evidence": {"local_material_quotes": 2},
+                "phase2_prompt_context": "章节：ch01\n必须围绕阶段门缺口求证。",
+            }
+        ],
+        "task_planning": {"phase2_mode": "chapter_integrated"},
+    }
+    framework = "## 第1章 审计发现应先转化为商业化阶段门缺口\n\n研究思路。"
+
+    cards = generate_task_cards_from_research_brief("baifan-test", framework, brief)
+
+    assert cards[0].prompt_brief == "章节：ch01\n必须围绕阶段门缺口求证。"
+    assert cards[0].research_goal == "本章要判断审计发现是否反映阶段门缺口。"
+    assert "大胆假设：风险项计数低估了商业化 readiness 缺口。" in cards[0].questions
+    assert cards[0].expected_evidence["phase1_minimum_evidence"] == {"local_material_quotes": 2}
+    assert cards[0].expected_evidence["must_address_phase1_hypothesis"] is True
+    assert any("Phase1 的大胆假设" in item for item in cards[0].stop_conditions)
+
+
+def test_phase1_gmp_hypotheses_are_not_title_restatements(tmp_path: Path) -> None:
+    project = tmp_path / "baifan"
+    project.mkdir()
+    (project / "phase0/extracted").mkdir(parents=True)
+    material = project / "phase0/extracted/audit.md"
+    material.write_text(
+        "人员培训记录齐全，但无菌操作动作违反 First Air 原则，需要进一步培训。\n"
+        "复盘显示 owner、关闭证据和问题升级机制仍需补齐。\n",
+        encoding="utf-8",
+    )
+    manifest = {
+        "topic": "白帆生物 GMP 与运营诊断",
+        "material_inventory": [{"extracted_to": "phase0/extracted/audit.md"}],
+    }
+    method = ResearchMethodRegistry().get("gmp_quality_operations_diagnosis")
+
+    plans = build_chapter_planning(
+        project,
+        manifest,
+        method,
+        ["人员能力：培训有效性比培训记录更关键", "运营节奏：从临时协调转向管理系统"],
+        quota=2000,
+    )
+
+    assert "关键解释变量" not in plans[0]["bold_hypothesis"]
+    assert "不缺培训台账" in plans[0]["bold_hypothesis"]
+    assert "固定节奏和可视化管理系统" in plans[1]["bold_hypothesis"]
+    assert plans[0]["core_question"] != plans[0]["title"]
+    assert "章节成稿应围绕这一观点展开" not in plans[1]["writing_claim"]
+
+
+def test_research_brief_without_materials_falls_back_to_material_digest() -> None:
+    brief = {
+        "research_method": "mckinsey_market",
+        "phase2_mode": "chapter_integrated",
+        "phase1_inputs": {"material_digest": "phase1/material_digest.md"},
+    }
+    framework = "## 第1章 市场需求正在被支付政策重塑\n\n研究思路。"
+
+    cards = generate_task_cards_from_research_brief("market-test", framework, brief)
+
+    assert cards[0].allowed_materials == ["phase1/material_digest.md"]
+
+
 def test_task_card_validation_rejects_duplicates_and_cycles() -> None:
     cards = [
         TaskCard(task_id="a", chapter_ids=["ch01"], topic_axis="clinical", questions=["q"], search_routes=["scholar"], output_packet="phase2/packets/a.json", dependencies=["b"]),
@@ -154,6 +273,10 @@ def test_packet_validation_requires_sources_and_counter_evidence() -> None:
         validate_packet(packet)
 
     packet["source_ids"].append("src_002")
+    packet["sources"] = [
+        {"id": "src_001", "title": "来源1", "url": "https://example.com/1"},
+        {"id": "src_002", "title": "来源2", "url": "https://example.com/2"},
+    ]
     validate_packet(packet)
 
 

tests/test_v020_workers.py

@@ -74,6 +74,10 @@ def valid_response() -> dict:
         "evidence_items": [{"source_id": "src_001", "summary": "III 期结果支持主要终点。"}],
         "counter_evidence": [{"claim": "长期安全性仍需随访", "source_ids": ["src_002"]}],
         "source_ids": ["src_001", "src_002"],
+        "sources": [
+            {"id": "src_001", "title": "来源1", "url": "https://example.com/1"},
+            {"id": "src_002", "title": "来源2", "url": "https://example.com/2"},
+        ],
         "source_quality_notes": ["src_001 Tier 1; src_002 Tier 2"],
         "open_questions": [],
         "raw_quotes_or_notes": ["Original English evidence note is allowed."],
@@ -98,6 +102,7 @@ def test_packet_worker_generates_valid_packet_with_fake_client(tmp_path: Path) -
 
     validate_packet(packet)
     assert fake.calls[0]["model"] == role.model
+    assert "章节证据分析师" in fake.calls[0]["system"]
     assert "search-strategy" in fake.calls[0]["system"]