merge: dev -> master（Daily Push + MinerU v4 + README sync）

.gitignore

@@ -7,6 +7,11 @@ __pycache__/
 env/
 venv/
 .env
+.env.*
+**/.env
+**/.env.*
+!.env.example
+!**/.env.example
 .venv
 build/
 dist/

README.md

@@ -19,7 +19,7 @@
 | **文献卡片** | Structured Card（LLM 提取 method/dataset/conclusion/limitations），懒加载 + DB 缓存 |
 | **导出增强** | BibTeX/RIS/Markdown/CSL-JSON（Zotero 原生导入），Next.js proxy route 修复 |
 | **写作辅助** | Related Work 草稿生成（基于 saved papers + topic），[AuthorYear] 引用格式，一键复制 |
-| **每日推送** | DailyPaper 生成后自动推送摘要到 Email/Slack/钉钉，支持 API 手动触发和 ARQ Cron 定时推送；MinerU 图表提取与 Apprise 多渠道（Telegram/Discord/企业微信/飞书/RSS）待集成 |
+| **每日推送** | DailyPaper 生成后自动推送到 Email/Slack/钉钉（支持 API 手动触发 + ARQ Cron 定时）；已集成 MinerU v4 主方法图提取（公网 URL + inline data URL 回退）、LLM Digest + Judge 评分卡片、Apprise 多渠道（Telegram/Discord/企业微信/飞书/RSS） |
 | **Model Provider** | 多 LLM 提供商管理（OpenAI/Anthropic/OpenRouter/Ollama），API Key Keychain 安全存储，任务级路由，连接测试 |
 | **Deadline Radar** | 会议截止日期追踪，CCF 分级过滤，Research Track 关键词匹配 |
 | **论文发现** | 种子论文扩展（引用/被引/共作者），Discovery Graph 可视化，论文集合（Collections）管理 |
@@ -35,7 +35,7 @@
 | DailyPaper | ✅ 可用 | `/research/paperscool/daily` | `daily-paper` | 报告生成 + LLM 增强 + Judge + 保存，完整可用 |
 | LLM-as-Judge | ✅ 可用 | `/research/paperscool/analyze` | `--with-judge` | 5 维评分 + 多轮校准 + 推荐分级 + Token Budget，SSE 增量推送 |
 | Analyze SSE | ✅ 可用 | `/research/paperscool/analyze` | — | Judge / Trend / Insight 三通道 SSE 流式，前端逐卡片渲染 |
-| Push/Notify | 🟡 基本可用 | `/research/paperscool/daily` | `--notify` | Email/Slack/钉钉 已落地；Apprise 多渠道（Telegram/Discord/企业微信/飞书/RSS）+ MinerU 图表提取待集成 |
+| Push/Notify | ✅ 可用 | `/research/paperscool/daily` | `--notify` | Email/Slack/钉钉 + Apprise 多渠道（Telegram/Discord/企业微信/飞书/RSS）；MinerU v4 主方法图（含 inline 回退）+ LLM Digest/Judge 卡片已落地 |
 | 学者追踪 | 🟡 基本可用 | `/track` | `track` | 多 Agent 管线 + PIS 评分完整；依赖 Semantic Scholar API Key |
 | 深度评审 | 🟡 基本可用 | `/review` | `review` | 模拟同行评审流程完整；输出质量取决于 LLM 后端配置 |
 | Paper2Code | 🟡 基本可用 | `/gen-code`（兼容） + `/research/repro/context/*` | `gen-code` | 编排 + RAG + CodeMemory 完整；执行层计划迁移为 AgentSwarm/Codex 专业执行器 |
@@ -172,7 +172,9 @@ Search → Build Report → LLM Enrichment → Judge Scoring → Filter → Save
 
 ### Email 推送
 
-![Email Notification](asset/notify.png)
+![Email Notification](asset/ui/dailypaperdemo.png)
+
+示例模板包含：导读摘要、Must Read 分层、Judge 评分、Digest Card、主方法图（可公开 URL 或 inline data URL 回退）。
 
 ## 快速开始
 
@@ -463,7 +465,12 @@ DB 持久化（统一主数据模型 Paper/Scholar/Event/Run）、任务队列/
 
 ### Phase 6 — 每日推送优化
 
-MinerU PDF 图表提取（主方法图自动识别）、推送内容增强（一句话总结 + 结构化摘要）、Apprise 多渠道统一推送层（Telegram/Discord/企业微信/飞书/RSS）、HuggingFace Daily Papers API 数据源接入。
+已完成：
+- MinerU v4 官方任务流接入（`/extract/task` + 轮询任务结果），主方法图自动识别。
+- 邮件渲染增强：支持主方法图展示（公开 URL）与 `data:image/...` inline 回退（适配 zip 内部图）。
+- 推送内容增强：导读摘要 + 一句话总结 + Digest Card + Judge 评分信息。
+- Apprise 多渠道统一推送层（Telegram/Discord/企业微信/飞书/RSS）落地。
+- HuggingFace Daily Papers API 数据源接入并纳入 DailyPaper 流程。
 
 ## 文档索引
 

env.example

@@ -27,6 +27,14 @@ OPENAI_BASE_URL=
 # ----------------------------
 SEMANTIC_SCHOLAR_API_KEY=
 GITHUB_TOKEN=
+MINERU_API_KEY=
+# MinerU Cloud API v4 async task endpoint
+MINERU_API_BASE_URL=https://mineru.net/api/v4
+# MinerU model_version, e.g. vlm / pipeline
+MINERU_MODEL_VERSION=vlm
+# Max wait for task polling in seconds
+MINERU_MAX_WAIT_SECONDS=180
+# MinerU Cloud limits: URL input only, <=200MB, <=600 pages, github/aws URL may timeout
 
 # CCS download (optional, ACM access URL)
 ACM_LIBRARY_URL=
@@ -101,6 +109,8 @@ PAPERBOT_DAILYPAPER_ENABLE_JUDGE=true
 PAPERBOT_DAILYPAPER_JUDGE_RUNS=1
 PAPERBOT_DAILYPAPER_JUDGE_MAX_ITEMS=5
 PAPERBOT_DAILYPAPER_JUDGE_TOKEN_BUDGET=0
+PAPERBOT_DAILYPAPER_ENABLE_FIGURES=false
+PAPERBOT_DAILYPAPER_FIGURES_MAX_ITEMS=5
 
 # ----------------------------
 # Daily push notifications

src/paperbot/api/routes/repro_context.py

@@ -23,6 +23,7 @@
 from paperbot.api.streaming import StreamEvent, wrap_generator
 from paperbot.application.services.p2c.models import (
     GenerateContextRequest as P2CRequest,
+    RawPaperData,
     new_context_pack_id,
 )
 from paperbot.application.services.p2c.orchestrator import ExtractionOrchestrator
@@ -44,6 +45,9 @@ class GenerateContextPackRequest(BaseModel):
     project_id: Optional[str] = None
     track_id: Optional[int] = None
     depth: Literal["fast", "standard", "deep"] = "standard"
+    # Optional inline paper data — when provided, skips the input router lookup.
+    title: Optional[str] = None
+    abstract: Optional[str] = None
 
 
 class CreateSessionRequest(BaseModel):
@@ -65,17 +69,23 @@ async def _generate_stream(request: GenerateContextPackRequest):
     )
 
     # Persist initial "running" record so the frontend can poll status.
-    await asyncio.to_thread(
-        _store.save,
-        pack_id=pack_id,
-        user_id=request.user_id,
-        paper_id=request.paper_id,
-        depth=request.depth,
-        pack_data={},
-        project_id=request.project_id,
-        confidence_overall=0.0,
-        warning_count=0,
-    )
+    try:
+        await asyncio.to_thread(
+            _store.save,
+            pack_id=pack_id,
+            user_id=request.user_id,
+            paper_id=request.paper_id,
+            depth=request.depth,
+            pack_data={},
+            project_id=request.project_id,
+            confidence_overall=0.0,
+            warning_count=0,
+        )
+    except Exception as exc:
+        Logger.warning(
+            f"[M2] store_save_failed pack_id={pack_id} error={exc} (continuing without persistence)",
+            file=LogFiles.ERROR,
+        )
 
     yield StreamEvent(type="status", data={"pack_id": pack_id, "status": "running"})
 
@@ -107,15 +117,22 @@ async def on_stage_complete(stage_name: str, observations: list, warnings: list)
             }
             for o in observations
         ]
-        await asyncio.to_thread(
-            _store.save_stage_result,
-            pack_id=pack_id,
-            stage_name=stage_name,
-            status="completed",
-            result_data={"observations": [o.to_full() for o in observations]},
-            confidence=confidence,
-            duration_ms=0,
-        )
+        try:
+            await asyncio.to_thread(
+                _store.save_stage_result,
+                pack_id=pack_id,
+                stage_name=stage_name,
+                status="completed",
+                result_data={"observations": [o.to_full() for o in observations]},
+                confidence=confidence,
+                duration_ms=0,
+            )
+        except Exception as exc:
+            Logger.warning(
+                f"[M2] store_save_stage_result_failed pack_id={pack_id} "
+                f"stage={stage_name} error={exc}",
+                file=LogFiles.ERROR,
+            )
         Logger.info(
             f"[M2] stage_complete pack_id={pack_id} stage={stage_name} obs={len(observations)} warnings={len(warnings)}",
             file=LogFiles.API,
@@ -150,9 +167,24 @@ async def on_stage_complete(stage_name: str, observations: list, warnings: list)
     orchestrator = ExtractionOrchestrator()
     result_holder: list = []
 
+    # When title+abstract are provided inline, build RawPaperData directly
+    # so the orchestrator skips the input router (which cannot handle studio IDs).
+    inline_raw: Optional[RawPaperData] = None
+    if request.title is not None and request.abstract is not None:
+        inline_raw = RawPaperData(
+            paper_id=request.paper_id,
+            title=request.title,
+            abstract=request.abstract,
+            source_adapter="inline",
+        )
+
     async def _run() -> None:
         try:
-            pack = await orchestrator.run(p2c_request, on_stage_complete=on_stage_complete)
+            pack = await orchestrator.run(
+                p2c_request,
+                raw_paper=inline_raw,
+                on_stage_complete=on_stage_complete,
+            )
             result_holder.append(pack)
             await queue.put(_DONE)
         except ValueError as exc:
@@ -175,8 +207,16 @@ async def _run() -> None:
             break
         elif item is _ERROR:
             err = result_holder[0]
-            await asyncio.to_thread(_store.update_status, pack_id, status="failed")
-            yield StreamEvent(type="error", data=err)
+            err_message = err.get("message", "Generation failed") if isinstance(err, dict) else str(err)
+            try:
+                await asyncio.to_thread(_store.update_status, pack_id, status="failed")
+            except Exception as exc:
+                Logger.warning(
+                    f"[M2] store_update_status_failed pack_id={pack_id} "
+                    f"status=failed error={exc}",
+                    file=LogFiles.ERROR,
+                )
+            yield StreamEvent(type="error", data=err, message=err_message)
             return
         else:
             yield item  # StreamEvent from on_stage_complete
@@ -196,15 +236,21 @@ async def _run() -> None:
     pack_dict = _asdict(pack)
     pack_dict["context_pack_id"] = pack_id  # align with our DB record
 
-    await asyncio.to_thread(
-        _store.update_status,
-        pack_id,
-        status="completed",
-        pack_data=pack_dict,
-        confidence_overall=pack.confidence.overall,
-        warning_count=len(pack.warnings),
-        objective=pack.objective,
-    )
+    try:
+        await asyncio.to_thread(
+            _store.update_status,
+            pack_id,
+            status="completed",
+            pack_data=pack_dict,
+            confidence_overall=pack.confidence.overall,
+            warning_count=len(pack.warnings),
+            objective=pack.objective,
+        )
+    except Exception as exc:
+        Logger.warning(
+            f"[M2] store_update_failed pack_id={pack_id} error={exc}",
+            file=LogFiles.ERROR,
+        )
     Logger.info(
         f"[M2] generation_completed pack_id={pack_id} observations={len(pack.observations)} warnings={len(pack.warnings)}",
         file=LogFiles.API,

src/paperbot/api/routes/runbook.py

@@ -155,13 +155,15 @@ def _normalize_user_directory(raw_path: str, field_name: str) -> Path:
 
     if not os.path.isabs(normalized):
         normalized = str((Path.cwd() / normalized).resolve(strict=False))
+    normalized_real = os.path.realpath(normalized)
 
     for prefix in _allowed_workdir_prefixes():
         prefix_str = str(prefix)
-        if normalized == prefix_str:
+        prefix_real = os.path.realpath(prefix_str)
+        if normalized_real == prefix_real:
             return prefix
-        if normalized.startswith(prefix_str + os.sep):
-            suffix = normalized[len(prefix_str):].lstrip("/\\")
+        if normalized_real.startswith(prefix_real + os.sep):
+            suffix = normalized_real[len(prefix_real):].lstrip("/\\")
             candidate = (prefix / suffix).resolve(strict=False) if suffix else prefix
             if _is_under_prefix(candidate, prefix):
                 return candidate
@@ -231,15 +233,26 @@ async def add_allowed_dir(body: AddAllowedDirRequest):
             detail="runtime allowlist mutation is disabled"
         )
 
-    try:
-        resolved = _normalize_user_directory(body.directory, field_name="directory")
-    except HTTPException:
-        raise
+    raw = body.directory.strip()
+    if not raw or "\x00" in raw:
+        raise HTTPException(status_code=400, detail="invalid directory path")
+
+    # Normalize and validate against configured safe prefixes before touching FS.
+    # This prevents path-injection style access to arbitrary locations.
+    resolved = _normalize_user_directory(raw, field_name="directory")
+
+    denied_roots = {Path(denied).resolve() for denied in _DENIED_PATHS}
+    if resolved in denied_roots:
+        raise HTTPException(
+            status_code=403,
+            detail=f"adding '{resolved}' is not allowed — path is too broad or sensitive",
+        )
 
-    if str(resolved) in _DENIED_PATHS:
+    # Deny home directory itself — too broad
+    if resolved == Path.home().resolve():
         raise HTTPException(
             status_code=403,
-            detail=f"adding '{resolved}' is not allowed — path is too broad or sensitive"
+            detail="adding home directory is not allowed — too broad",
         )
 
     if not resolved.exists() or not resolved.is_dir():