deepresearch-flow 0.4.1__py3-none-any.whl → 0.5.1__py3-none-any.whl

deepresearch_flow/paper/prompt_templates/simple_phi_user.j2

@@ -0,0 +1,31 @@
+Template: simple_phi
+
+{% if stage_name %}
+Current stage: {{ stage_name }}.
+Return JSON with keys: {{ stage_fields | join(", ") }}.
+Output language: {{ output_language }}. Use that language in all answers (if zh, use Chinese).
+Previously completed outputs (JSON): {{ previous_outputs }}
+{% else %}
+你是“哲学/伦理学论文高密度摘要写手”。基于我提供的论文全文/摘录/笔记，写一段单段摘要（900–1100字，禁止第一人称），风格参考：先讲问题与争点→指出现有立场/方法的不足→给出本文主论题→用3–6句复述论证主链（按“首先/其次/最后”串起来）→点出2–4个关键概念/区分（每个用极短句说明其在文中的用法/定义）→交代作者处理的1–2条关键反对意见及回应→用2–3句总结贡献/意义→用1–2句写局限与后续方向。
+
+硬约束：
+1) 只写论文中明确支持的内容；超出原文必须标【推测】；论文没交代的证据/覆盖范围写“论文未给出”。
+2) 若可能，在段内插入≥5个最小定位标签（例如{Sect.X.Y}或{锚点:段首8-12字}）。若无法定位，标【未定位】并说明原因（材料缺页码/无小节/摘录不完整等）。
+3) 不输出列表/分段/小标题，只输出一整段连续文本（允许少量分号）。
+4) 输出语言：{{ output_language }}。请使用该语言输出（zh 时使用中文，保留必要英文术语）。
+
+段末必须追加一行“核心压缩表达（四选一）”，从论文中抽取最关键的一条，以纯ASCII表示、避免引号和特殊符号：
+- 核心判准：X iff (A and B and C)
+- 或 核心定义：X := (R1,R2,R3)
+- 或 核心原则：If P then Q
+- 或 核心论证骨架：P1 + P2 + ... + Pn => C
+选择规则：若论文给出明确必要/充分条件或工作定义，优先用“判准/定义”；若是规范伦理，优先用“原则”；若不存在单一判准/原则，则用“论证骨架”。
+
+现在开始输出这一段摘要。
+{% endif %}
+
+Document content:
+{{ content }}
+
+JSON Schema:
+{{ schema }}

deepresearch_flow/paper/schemas/deep_read_phi_schema.json

@@ -0,0 +1,30 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "type": "object",
+  "additionalProperties": false,
+  "required": [
+    "paper_title",
+    "paper_authors",
+    "publication_date",
+    "publication_venue",
+    "module_m1",
+    "module_m2",
+    "module_m3",
+    "module_m4",
+    "module_m5",
+    "module_m6"
+  ],
+  "properties": {
+    "paper_title": {"type": "string", "minLength": 1},
+    "paper_authors": {"type": "array", "items": {"type": "string"}},
+    "publication_date": {"type": "string"},
+    "publication_venue": {"type": "string"},
+    "module_m1": {"type": "string"},
+    "module_m2": {"type": "string"},
+    "module_m3": {"type": "string"},
+    "module_m4": {"type": "string"},
+    "module_m5": {"type": "string"},
+    "module_m6": {"type": "string"},
+    "output_language": {"type": "string"}
+  }
+}

deepresearch_flow/paper/template_registry.py

@@ -34,6 +34,13 @@ _TEMPLATES: dict[str, TemplateBundle] = {
         schema_file="default_paper_schema.json",
         render_template="default_paper.md.j2",
     ),
+    "simple_phi": TemplateBundle(
+        name="simple_phi",
+        prompt_system="simple_phi_system.j2",
+        prompt_user="simple_phi_user.j2",
+        schema_file="default_paper_schema.json",
+        render_template="default_paper.md.j2",
+    ),
     "deep_read": TemplateBundle(
         name="deep_read",
         prompt_system="deep_read_system.j2",
@@ -41,6 +48,13 @@ _TEMPLATES: dict[str, TemplateBundle] = {
         schema_file="deep_read_schema.json",
         render_template="deep_read.md.j2",
     ),
+    "deep_read_phi": TemplateBundle(
+        name="deep_read_phi",
+        prompt_system="deep_read_phi_system.j2",
+        prompt_user="deep_read_phi_user.j2",
+        schema_file="deep_read_phi_schema.json",
+        render_template="deep_read_phi.md.j2",
+    ),
     "eight_questions": TemplateBundle(
         name="eight_questions",
         prompt_system="eight_questions_system.j2",
@@ -48,6 +62,13 @@ _TEMPLATES: dict[str, TemplateBundle] = {
         schema_file="eight_questions_schema.json",
         render_template="eight_questions.md.j2",
     ),
+    "eight_questions_phi": TemplateBundle(
+        name="eight_questions_phi",
+        prompt_system="eight_questions_phi_system.j2",
+        prompt_user="eight_questions_phi_user.j2",
+        schema_file="eight_questions_schema.json",
+        render_template="eight_questions.md.j2",
+    ),
     "three_pass": TemplateBundle(
         name="three_pass",
         prompt_system="three_pass_system.j2",
@@ -75,6 +96,14 @@ _STAGES: dict[str, list[StageDefinition]] = {
         StageDefinition("module_g", ["module_g"]),
         StageDefinition("module_h", ["module_h"]),
     ],
+    "deep_read_phi": [
+        StageDefinition("module_m1", ["module_m1"]),
+        StageDefinition("module_m2", ["module_m2"]),
+        StageDefinition("module_m3", ["module_m3"]),
+        StageDefinition("module_m4", ["module_m4"]),
+        StageDefinition("module_m5", ["module_m5"]),
+        StageDefinition("module_m6", ["module_m6"]),
+    ],
     "eight_questions": [
         StageDefinition(
             "questions_1to4",
@@ -85,6 +114,16 @@ _STAGES: dict[str, list[StageDefinition]] = {
             ["question5", "question6", "question7", "question8"],
         ),
     ],
+    "eight_questions_phi": [
+        StageDefinition(
+            "questions_1to4",
+            ["question1", "question2", "question3", "question4"],
+        ),
+        StageDefinition(
+            "questions_5to8",
+            ["question5", "question6", "question7", "question8"],
+        ),
+    ],
     "three_pass": [
         StageDefinition("step1_summary", ["step1_summary"]),
         StageDefinition("step2_analysis", ["step2_analysis"]),

deepresearch_flow/paper/templates/deep_read_phi.md.j2

@@ -0,0 +1,40 @@
+# {{ paper_title }}
+
+{% set is_zh = output_language == "zh" %}
+**{{ "作者 / Authors" if is_zh else "Authors" }}:** {{ paper_authors | join(", ") }}
+
+{% if output_language %}
+**{{ "输出语言 / Output Language" if is_zh else "Output Language" }}:** {{ output_language }}
+{% endif %}
+
+{% if publication_date %}
+**{{ "发表日期 / Publication Date" if is_zh else "Publication Date" }}:** {{ publication_date }}
+{% endif %}
+
+{% if publication_venue %}
+**{{ "期刊/会议 / Publication Venue" if is_zh else "Publication Venue" }}:** {{ publication_venue }}
+{% endif %}
+
+## {{ "模块 M1：对齐目标 + 输入校验 + 论文地图 + Exhibit全量索引 + markmap大纲脑图" if is_zh else "Module M1: Alignment + Input Check + Paper Map + Exhibit Index + Markmap" }}
+
+{{ module_m1 }}
+
+## {{ "模块 M2：第一遍鸟瞰 + Exhibit客观打分排序" if is_zh else "Module M2: First Pass Overview + Exhibit Scoring" }}
+
+{{ module_m2 }}
+
+## {{ "模块 M3：概念与术语工程 + 辩论谱系" if is_zh else "Module M3: Concepts + Debate Lineage" }}
+
+{{ module_m3 }}
+
+## {{ "模块 M4：论证重建 + Top Exhibits 深读（上半）" if is_zh else "Module M4: Argument Map + Top Exhibits (Part 1)" }}
+
+{{ module_m4 }}
+
+## {{ "模块 M5：深度审视 + objection mining + Top Exhibits 深读（下半）" if is_zh else "Module M5: Deep Review + Objection Mining + Top Exhibits (Part 2)" }}
+
+{{ module_m5 }}
+
+## {{ "模块 M6：写作级产出包" if is_zh else "Module M6: Writing-Ready Output Pack" }}
+
+{{ module_m6 }}

deepresearch_flow/paper/web/app.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import logging
+import os
 from pathlib import Path
 
 from starlette.applications import Starlette
@@ -10,8 +11,9 @@ from starlette.routing import Mount, Route
 from starlette.staticfiles import StaticFiles
 
 from deepresearch_flow.paper.db_ops import build_index, load_and_merge_papers
-from deepresearch_flow.paper.web.constants import PDFJS_STATIC_DIR, STATIC_DIR
+from deepresearch_flow.paper.web.constants import DEFAULT_PDFJS_CDN_BASE_URL, PDFJS_STATIC_DIR, STATIC_DIR
 from deepresearch_flow.paper.web.handlers import (
+    api_markdown,
     api_papers,
     api_pdf,
     api_stats,
@@ -21,6 +23,7 @@ from deepresearch_flow.paper.web.handlers import (
     stats_page,
 )
 from deepresearch_flow.paper.web.markdown import create_md_renderer
+from deepresearch_flow.paper.web.static_assets import build_static_assets
 
 logger = logging.getLogger(__name__)
 
@@ -32,6 +35,35 @@ class _NoIndexMiddleware(BaseHTTPMiddleware):
         return response
 
 
+class _StaticAssetFiles(StaticFiles):
+    def __init__(self, *args, cache_control: str | None = None, **kwargs) -> None:
+        super().__init__(*args, **kwargs)
+        self._cache_control = cache_control
+
+    async def get_response(self, path: str, scope):  # type: ignore[override]
+        response = await super().get_response(path, scope)
+        if self._cache_control and response.status_code == 200:
+            response.headers.setdefault("Cache-Control", self._cache_control)
+        return response
+
+
+def _normalize_static_mode(value: str | None) -> str:
+    if not value:
+        return "auto"
+    normalized = value.strip().lower()
+    if normalized in {"dev", "development"}:
+        return "dev"
+    if normalized in {"prod", "production"}:
+        return "prod"
+    return "auto"
+
+
+def _resolve_static_mode(value: str, static_base_url: str | None) -> str:
+    if value == "auto":
+        return "prod" if static_base_url else "dev"
+    return value
+
+
 def create_app(
     *,
     db_paths: list[Path],
@@ -42,6 +74,10 @@ def create_app(
     pdf_roots: list[Path] | None = None,
     cache_dir: Path | None = None,
     use_cache: bool = True,
+    static_base_url: str | None = None,
+    static_mode: str | None = None,
+    static_export_dir: Path | None = None,
+    pdfjs_cdn_base_url: str | None = None,
 ) -> Starlette:
     papers = load_and_merge_papers(db_paths, bibtex_path, cache_dir, use_cache, pdf_roots=pdf_roots)
 
@@ -55,6 +91,44 @@ def create_app(
         pdf_roots=pdf_roots,
     )
     md = create_md_renderer()
+    static_base_url = static_base_url or os.getenv("PAPER_DB_STATIC_BASE_URL")
+    static_mode = _normalize_static_mode(static_mode or os.getenv("PAPER_DB_STATIC_MODE"))
+    resolved_mode = _resolve_static_mode(static_mode, static_base_url)
+    export_dir_value = static_export_dir or os.getenv("PAPER_DB_STATIC_EXPORT_DIR")
+    export_dir = Path(export_dir_value) if export_dir_value else None
+    pdfjs_cdn_base_url = (
+        pdfjs_cdn_base_url
+        or os.getenv("PAPER_DB_PDFJS_CDN_BASE_URL")
+        or DEFAULT_PDFJS_CDN_BASE_URL
+    )
+    if pdfjs_cdn_base_url:
+        lowered = pdfjs_cdn_base_url.strip().lower()
+        if lowered in {"none", "off", "local"}:
+            pdfjs_cdn_base_url = None
+        else:
+            pdfjs_cdn_base_url = pdfjs_cdn_base_url.rstrip("/")
+
+    asset_config = None
+    if resolved_mode == "prod":
+        if not static_base_url:
+            logger.warning("Static mode set to prod without base URL; falling back to dev asset routes.")
+            resolved_mode = "dev"
+        else:
+            asset_config = build_static_assets(
+                index,
+                static_base_url=static_base_url,
+                static_export_dir=export_dir,
+            )
+    if resolved_mode == "dev" and export_dir:
+        asset_config = build_static_assets(
+            index,
+            static_base_url="",
+            static_export_dir=export_dir,
+            allow_empty_base=True,
+        )
+    if asset_config is None:
+        asset_config = build_static_assets(index, static_base_url=None)
+
     routes = [
         Route("/", index_page, methods=["GET"]),
         Route("/robots.txt", robots_txt, methods=["GET"]),
@@ -63,6 +137,7 @@ def create_app(
         Route("/api/papers", api_papers, methods=["GET"]),
         Route("/api/stats", api_stats, methods=["GET"]),
         Route("/api/pdf/{source_hash:str}", api_pdf, methods=["GET"]),
+        Route("/api/dev/markdown/{source_hash:str}", api_markdown, methods=["GET"]),
     ]
     if PDFJS_STATIC_DIR.exists():
         routes.append(
@@ -85,10 +160,40 @@ def create_app(
                 name="static",
             )
         )
+    if export_dir and export_dir.exists() and asset_config.enabled and not asset_config.base_url:
+        cache_header = "public, max-age=31536000, immutable"
+        routes.extend(
+            [
+                Mount(
+                    "/pdf",
+                    app=_StaticAssetFiles(directory=str(export_dir / "pdf"), cache_control=cache_header),
+                    name="static_pdf",
+                ),
+                Mount(
+                    "/images",
+                    app=_StaticAssetFiles(directory=str(export_dir / "images"), cache_control=cache_header),
+                    name="static_images",
+                ),
+                Mount(
+                    "/md",
+                    app=_StaticAssetFiles(directory=str(export_dir / "md"), cache_control=cache_header),
+                    name="static_md",
+                ),
+                Mount(
+                    "/md_translate",
+                    app=_StaticAssetFiles(directory=str(export_dir / "md_translate"), cache_control=cache_header),
+                    name="static_md_translate",
+                ),
+            ]
+        )
     app = Starlette(routes=routes)
     app.add_middleware(_NoIndexMiddleware)
     app.state.index = index
     app.state.md = md
     app.state.fallback_language = fallback_language
     app.state.pdf_roots = pdf_roots
+    app.state.static_mode = resolved_mode
+    app.state.asset_config = asset_config
+    app.state.static_export_dir = export_dir
+    app.state.pdfjs_cdn_base_url = pdfjs_cdn_base_url
     return app

deepresearch_flow/paper/web/constants.py

@@ -12,6 +12,7 @@ CDN_KATEX_AUTO = "https://cdn.jsdelivr.net/npm/katex@0.16.27/dist/contrib/auto-r
 # Use legacy builds to ensure `pdfjsLib` is available as a global.
 CDN_PDFJS = "https://cdn.jsdelivr.net/npm/pdfjs-dist@3.11.174/legacy/build/pdf.min.js"
 CDN_PDFJS_WORKER = "https://cdn.jsdelivr.net/npm/pdfjs-dist@3.11.174/legacy/build/pdf.worker.min.js"
+DEFAULT_PDFJS_CDN_BASE_URL = "https://cdn.jsdelivr.net/npm/pdfjs-dist@3.11.174"
 
 # PDF.js viewer configuration
 PDFJS_VIEWER_PATH = "/pdfjs/web/viewer.html"

deepresearch_flow/paper/web/handlers/__init__.py

@@ -1,12 +1,13 @@
 """Route handlers for paper web UI."""
 
-from .api import api_papers, api_pdf, api_stats
+from .api import api_markdown, api_papers, api_pdf, api_stats
 from .pages import index_page, paper_detail, robots_txt, stats_page
 
 __all__ = [
     "api_papers",
     "api_pdf",
     "api_stats",
+    "api_markdown",
     "index_page",
     "paper_detail",
     "robots_txt",

deepresearch_flow/paper/web/handlers/api.py

@@ -19,6 +19,8 @@ from deepresearch_flow.paper.web.filters import (
     presence_filter,
     sorted_ids,
 )
+from deepresearch_flow.paper.web.markdown import normalize_markdown_images
+from deepresearch_flow.paper.web.static_assets import resolve_asset_urls
 from deepresearch_flow.paper.web.text import extract_summary_snippet, normalize_title, normalize_venue
 from deepresearch_flow.paper.web.query import Query, QueryTerm, parse_query
 
@@ -92,9 +94,18 @@ def _apply_query(index: PaperIndex, query: Query) -> set[int]:
     return result
 
 
+def _safe_read_text(path: Path) -> str:
+    try:
+        return path.read_text(encoding="utf-8")
+    except UnicodeDecodeError:
+        return path.read_text(encoding="latin-1")
+
+
 async def api_papers(request: Request) -> JSONResponse:
     """API endpoint for paper list with filtering, sorting, and pagination."""
     index: PaperIndex = request.app.state.index
+    asset_config = request.app.state.asset_config
+    prefer_local = request.app.state.static_mode == "dev"
     filters = parse_filters(request)
     page = int(filters["page"])
     page_size = int(filters["page_size"])
@@ -165,6 +176,7 @@ async def api_papers(request: Request) -> JSONResponse:
         source_hash = str(paper.get("source_hash") or stable_hash(str(paper.get("source_path") or idx)))
         translations = index.translated_md_by_hash.get(source_hash, {})
         translation_languages = sorted(translations.keys(), key=str.lower)
+        asset_urls = resolve_asset_urls(index, source_hash, asset_config, prefer_local=prefer_local)
         items.append(
             {
                 "source_hash": source_hash,
@@ -183,6 +195,10 @@ async def api_papers(request: Request) -> JSONResponse:
                 "has_summary": bool(paper.get("_has_summary")),
                 "is_pdf_only": bool(paper.get("_is_pdf_only")),
                 "translation_languages": translation_languages,
+                "pdf_url": asset_urls["pdf_url"],
+                "md_url": asset_urls["md_url"],
+                "md_translated_url": asset_urls["md_translated_url"],
+                "images_base_url": asset_urls["images_base_url"],
             }
         )
 
@@ -215,3 +231,42 @@ async def api_pdf(request: Request) -> Response:
     if allowed_roots and not _ensure_under_roots(pdf_path, allowed_roots):
         return Response("Forbidden", status_code=403)
     return FileResponse(pdf_path)
+
+
+async def api_markdown(request: Request) -> Response:
+    """Dev-only API endpoint to serve raw markdown content."""
+    if request.app.state.static_mode != "dev":
+        return Response("Not Found", status_code=404)
+    index: PaperIndex = request.app.state.index
+    asset_config = request.app.state.asset_config
+    export_dir = request.app.state.static_export_dir
+    source_hash = request.path_params["source_hash"]
+    lang = request.query_params.get("lang")
+    md_path = None
+    if export_dir and asset_config and asset_config.enabled and (asset_config.base_url or "") == "":
+        if lang:
+            translated_url = asset_config.translated_md_urls.get(source_hash, {}).get(lang.lower())
+            if translated_url:
+                rel_path = translated_url.lstrip("/")
+                export_path = export_dir / rel_path
+                if export_path.exists():
+                    raw = _safe_read_text(export_path)
+                    return Response(raw, media_type="text/markdown")
+        else:
+            md_url = asset_config.md_urls.get(source_hash)
+            if md_url:
+                rel_path = md_url.lstrip("/")
+                export_path = export_dir / rel_path
+                if export_path.exists():
+                    raw = _safe_read_text(export_path)
+                    return Response(raw, media_type="text/markdown")
+    if lang:
+        md_path = index.translated_md_by_hash.get(source_hash, {}).get(lang.lower())
+    else:
+        md_path = index.md_path_by_hash.get(source_hash)
+    if not md_path:
+        return Response("Markdown not found", status_code=404)
+    raw = _safe_read_text(md_path)
+    if lang:
+        raw = normalize_markdown_images(raw)
+    return Response(raw, media_type="text/markdown")