diff --git a/.gitignore b/.gitignore
index cf653b3..e593c84 100644
--- a/.gitignore
+++ b/.gitignore
@@ -17,13 +17,8 @@ build/
venv/
env/
-# Runtime outputs (keep result-template examples)
-data/result-*/
-!data/result-template1/
-!data/result-template2/
-
-# Runtime cache and config
-data/cache/
+# Runtime outputs and cache
+data/
config.json
.superpowers/
diff --git a/citationclaw/app/main.py b/citationclaw/app/main.py
index a19e869..665f212 100644
--- a/citationclaw/app/main.py
+++ b/citationclaw/app/main.py
@@ -530,6 +530,78 @@ async def cancel_task():
return {"status": "success", "message": "任务取消中..."}
+@app.post("/api/task/resume")
+async def resume_task():
+ """恢复被取消的任务"""
+ if task_executor.is_running:
+ return JSONResponse(status_code=400,
+ content={"status": "error", "message": "任务正在运行中"})
+ if not task_executor._resume_state:
+ return JSONResponse(status_code=400,
+ content={"status": "error", "message": "没有可恢复的任务"})
+
+ resume = task_executor._resume_state
+ config = config_manager.get()
+ task_executor.current_task = asyncio.create_task(
+ task_executor.execute_for_titles(
+ paper_groups=resume["paper_groups"],
+ config=config,
+ output_prefix=resume["output_prefix"],
+ _resume=resume,
+ )
+ )
+ return {
+ "status": "success",
+ "message": f"已恢复，已完成的阶段将自动跳过"
+ }
+
+
+class ResumeFolderRequest(BaseModel):
+ folder: str # e.g. "result-20260317_155742"
+
+
+@app.post("/api/task/resume-from-folder")
+async def resume_from_folder(request: ResumeFolderRequest):
+ """从结果文件夹恢复任务"""
+ import json as _json
+
+ if task_executor.is_running:
+ return JSONResponse(status_code=400,
+ content={"status": "error", "message": "任务正在运行中"})
+
+ result_dir = Path("data") / request.folder
+ if not result_dir.is_dir():
+ return JSONResponse(status_code=400,
+ content={"status": "error", "message": f"文件夹不存在: {request.folder}"})
+
+ meta_file = result_dir / "_task_meta.json"
+ if not meta_file.exists():
+ return JSONResponse(status_code=400,
+ content={"status": "error", "message": "该文件夹缺少任务元数据（_task_meta.json），无法恢复"})
+
+ try:
+ meta = _json.loads(meta_file.read_text(encoding="utf-8"))
+ except Exception as e:
+ return JSONResponse(status_code=400,
+ content={"status": "error", "message": f"读取元数据失败: {e}"})
+
+ config = config_manager.get()
+ resume = {
+ "paper_groups": meta["paper_groups"],
+ "output_prefix": meta["output_prefix"],
+ "result_dir": str(result_dir),
+ }
+ task_executor.current_task = asyncio.create_task(
+ task_executor.execute_for_titles(
+ paper_groups=resume["paper_groups"],
+ config=config,
+ output_prefix=resume["output_prefix"],
+ _resume=resume,
+ )
+ )
+ return {"status": "success", "message": f"已从 {request.folder} 恢复任务，已完成的阶段将自动跳过"}
+
+
@app.post("/api/task/year-traverse-respond")
async def year_traverse_respond(request: YearTraverseResponse):
"""接收用户对年份遍历提示的响应"""
@@ -633,6 +705,7 @@ async def list_result_folders():
"file_count": len(files),
"modified": max((f.stat().st_mtime for f in files), default=sub.stat().st_mtime),
"size": sum(f.stat().st_size for f in files),
+ "has_meta": (sub / "_task_meta.json").exists(),
})
# 旧版扁平目录
diff --git a/citationclaw/app/task_executor.py b/citationclaw/app/task_executor.py
index 4d9bf8b..8f0cf9f 100644
--- a/citationclaw/app/task_executor.py
+++ b/citationclaw/app/task_executor.py
@@ -31,6 +31,9 @@ def __init__(self, log_manager: LogManager):
self._year_traverse_prompted: bool = False # 本次运行已提示过，不再重复
self.skills_runtime = SkillsRuntime()
+ # 取消后可恢复的上下文
+ self._resume_state: Optional[dict] = None
+
async def _run_skill(self, skill_name: str, config: AppConfig, **kwargs):
"""Execute one pipeline skill with shared runtime context."""
return await self.skills_runtime.run(
@@ -133,6 +136,8 @@ async def execute_full_pipeline(
self.log_manager.success(f"JSON文件: {json_file}")
self.log_manager.success("=" * 50)
+ except asyncio.CancelledError:
+ self.log_manager.warning("任务已取消")
except Exception as e:
self.log_manager.error(f"任务执行错误: {str(e)}")
import traceback
@@ -140,6 +145,7 @@ async def execute_full_pipeline(
raise
finally:
self.is_running = False
+ self.should_cancel = False
async def execute_stage1_scraping(
self,
@@ -208,6 +214,8 @@ async def execute_stage1_scraping(
}
})
+ except asyncio.CancelledError:
+ self.log_manager.warning("任务已取消")
except Exception as e:
self.log_manager.error(f"阶段1执行错误: {str(e)}")
import traceback
@@ -215,6 +223,7 @@ async def execute_stage1_scraping(
raise
finally:
self.is_running = False
+ self.should_cancel = False
async def import_history(self, file_path: Path, config: AppConfig) -> dict:
"""
@@ -337,6 +346,8 @@ async def execute_stage2_and_3(self):
# 清空阶段1结果
self.stage1_result = None
+ except asyncio.CancelledError:
+ self.log_manager.warning("任务已取消")
except Exception as e:
self.log_manager.error(f"阶段2/3执行错误: {str(e)}")
import traceback
@@ -344,12 +355,14 @@ async def execute_stage2_and_3(self):
raise
finally:
self.is_running = False
+ self.should_cancel = False
async def execute_for_titles(
self,
paper_groups: List[dict],
config: AppConfig,
output_prefix: str,
+ _resume: Optional[dict] = None,
):
"""
全自动多论文流水线：
@@ -363,6 +376,7 @@ async def execute_for_titles(
self.is_running = True
self.should_cancel = False
+ self._resume_state = None
self._year_traverse_event = None
self._year_traverse_choice = False
self._year_traverse_prompted = False
@@ -372,10 +386,25 @@ async def execute_for_titles(
cost_tracker = CostTracker()
try:
- timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
- result_dir = Path(f"data/result-{timestamp}")
- result_dir.mkdir(parents=True, exist_ok=True)
- self.log_manager.info(f"📁 结果目录: {result_dir}")
+ # 恢复模式：复用之前的结果目录，靠文件存在性自动跳过已完成阶段
+ if _resume:
+ result_dir = Path(_resume["result_dir"])
+ self.log_manager.info(f"🔄 恢复任务，结果目录: {result_dir}（已完成的阶段将自动跳过）")
+ else:
+ timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+ result_dir = Path(f"data/result-{timestamp}")
+ result_dir.mkdir(parents=True, exist_ok=True)
+ self.log_manager.info(f"📁 结果目录: {result_dir}")
+ author_info_files = []
+ citing_files = []
+
+ # 保存任务元数据（用于从结果文件夹恢复）
+ meta_file = result_dir / "_task_meta.json"
+ if not meta_file.exists():
+ meta_file.write_text(_json.dumps({
+ "paper_groups": paper_groups,
+ "output_prefix": output_prefix,
+ }, ensure_ascii=False, indent=2), encoding="utf-8")
# 运行前快照 LLM 额度
self.log_manager.info(f"📊 [DEBUG] api_access_token={repr(config.api_access_token[:8] + '...' if config.api_access_token else '')}, api_user_id={repr(config.api_user_id)}")
@@ -401,8 +430,6 @@ async def execute_for_titles(
alias_to_canonical[alias] = canonical
all_search_titles.append((alias, canonical))
- author_info_files = [] # 收集每篇/每别名的 Phase 2 输出
- citing_files = [] # 收集每篇/每别名的 Phase 1 输出
total = len(all_search_titles)
# 目标论文作者缓存（每个 canonical 只查询一次 LLM，存完整文本）
@@ -464,97 +491,106 @@ async def execute_for_titles(
continue
paper_slug = f"paper{i+1}"
+ citing_file = result_dir / f"{paper_slug}_citing.jsonl"
+ author_file = result_dir / f"{paper_slug}_authors.jsonl"
- # —— Phase 1：爬取引用列表 ——
- self.log_manager.info("▶ Phase 1: 爬取引用列表")
+ # —— Phase 1：爬取引用列表（已有输出文件则跳过）——
+ if citing_file.exists() and citing_file.stat().st_size > 0:
+ self.log_manager.info(f"⏩ Phase 1 已完成，跳过: {citing_file.name}")
+ else:
+ self.log_manager.info("▶ Phase 1: 爬取引用列表")
+
+ # —— 预检测引用数，超1000时询问是否开启年份遍历 ——
+ if not config.enable_year_traverse and not self._year_traverse_prompted:
+ probe_data = await self._run_skill(
+ "phase1_citation_fetch",
+ config,
+ url=url,
+ probe_only=True,
+ cost_tracker=cost_tracker,
+ )
+ citation_count = int(probe_data.get("citation_count", 0))
+ if citation_count > 1000:
+ self._year_traverse_prompted = True
+ self._year_traverse_event = asyncio.Event()
+ self.log_manager.broadcast_event("year_traverse_prompt", {
+ "title": title,
+ "citation_count": citation_count,
+ })
+ self.log_manager.warning(
+ f"⚠️ 论文「{title}」检测到 {citation_count} 篇引用（超过 Google Scholar 1000 条限制）"
+ )
+ self.log_manager.warning(
+ "⏸ 已暂停，等待用户选择是否启用按年份遍历（最多等待 60 秒）..."
+ )
+ try:
+ await asyncio.wait_for(self._year_traverse_event.wait(), timeout=60.0)
+ if self._year_traverse_choice:
+ config = config.model_copy(update={"enable_year_traverse": True})
+ self.log_manager.info("✅ 已启用按年份遍历，将逐年抓取完整数据")
+ else:
+ self.log_manager.info("▶ 已跳过，继续普通模式（可能只抓取前 1000 条）")
+ except asyncio.TimeoutError:
+ self.log_manager.warning("⏰ 等待超时（60s），以普通模式继续")
+ finally:
+ self._year_traverse_event = None
- # —— 预检测引用数，超1000时询问是否开启年份遍历 ——
- if not config.enable_year_traverse and not self._year_traverse_prompted:
- probe_data = await self._run_skill(
+ await self._run_skill(
"phase1_citation_fetch",
config,
url=url,
- probe_only=True,
+ output_file=citing_file,
+ start_page=0,
+ sleep_seconds=config.sleep_between_pages,
+ enable_year_traverse=config.enable_year_traverse,
cost_tracker=cost_tracker,
)
- citation_count = int(probe_data.get("citation_count", 0))
- if citation_count > 1000:
- self._year_traverse_prompted = True
- self._year_traverse_event = asyncio.Event()
- self.log_manager.broadcast_event("year_traverse_prompt", {
- "title": title,
- "citation_count": citation_count,
- })
- self.log_manager.warning(
- f"⚠️ 论文「{title}」检测到 {citation_count} 篇引用（超过 Google Scholar 1000 条限制）"
- )
- self.log_manager.warning(
- "⏸ 已暂停，等待用户选择是否启用按年份遍历（最多等待 60 秒）..."
- )
- try:
- await asyncio.wait_for(self._year_traverse_event.wait(), timeout=60.0)
- if self._year_traverse_choice:
- config = config.model_copy(update={"enable_year_traverse": True})
- self.log_manager.info("✅ 已启用按年份遍历，将逐年抓取完整数据")
- else:
- self.log_manager.info("▶ 已跳过，继续普通模式（可能只抓取前 1000 条）")
- except asyncio.TimeoutError:
- self.log_manager.warning("⏰ 等待超时（60s），以普通模式继续")
- finally:
- self._year_traverse_event = None
+ if self.should_cancel:
+ break
- citing_file = result_dir / f"{paper_slug}_citing.jsonl"
- await self._run_skill(
- "phase1_citation_fetch",
- config,
- url=url,
- output_file=citing_file,
- start_page=0,
- sleep_seconds=config.sleep_between_pages,
- enable_year_traverse=config.enable_year_traverse,
- cost_tracker=cost_tracker,
- )
- if self.should_cancel:
- break
citing_files.append((citing_file, canonical))
if not config.skip_author_search:
- # —— 获取目标论文作者（用于自引检测，每个 canonical 只查一次）——
- if canonical not in target_authors_cache:
- need_self_filter = (
- config.enable_renowned_scholar_filter or config.enable_citing_description
+ # —— Phase 2：搜索作者信息（已有输出文件则跳过）——
+ if author_file.exists() and author_file.stat().st_size > 0:
+ self.log_manager.info(f"⏩ Phase 2 已完成，跳过: {author_file.name}")
+ author_info_files.append(author_file)
+ else:
+ # —— 获取目标论文作者（用于自引检测，每个 canonical 只查一次）——
+ if canonical not in target_authors_cache:
+ need_self_filter = (
+ config.enable_renowned_scholar_filter or config.enable_citing_description
+ )
+ if not config.test_mode and need_self_filter:
+ self.log_manager.info("🔍 自引检测：正在获取目标论文作者...")
+ target_authors_cache[canonical] = await self._fetch_target_authors(canonical, config)
+ else:
+ target_authors_cache[canonical] = ""
+ target_authors = target_authors_cache[canonical]
+
+ self.log_manager.info("▶ Phase 2: 搜索作者信息")
+ await self._run_skill(
+ "phase2_author_intel",
+ config,
+ input_file=citing_file,
+ output_file=author_file,
+ sleep_seconds=config.sleep_between_authors,
+ parallel_workers=config.parallel_author_search,
+ citing_paper=canonical, # 始终用正式标题写入 Citing_Paper
+ target_paper_authors=target_authors,
+ author_cache=author_cache,
+ quota_event=self.quota_exceeded_event,
)
- if not config.test_mode and need_self_filter:
- self.log_manager.info("🔍 自引检测：正在获取目标论文作者...")
- target_authors_cache[canonical] = await self._fetch_target_authors(canonical, config)
- else:
- target_authors_cache[canonical] = ""
- target_authors = target_authors_cache[canonical]
-
- # —— Phase 2：搜索作者信息（以 canonical 为 Citing_Paper 值）——
- self.log_manager.info("▶ Phase 2: 搜索作者信息")
- author_file = result_dir / f"{paper_slug}_authors.jsonl"
- await self._run_skill(
- "phase2_author_intel",
- config,
- input_file=citing_file,
- output_file=author_file,
- sleep_seconds=config.sleep_between_authors,
- parallel_workers=config.parallel_author_search,
- citing_paper=canonical, # 始终用正式标题写入 Citing_Paper
- target_paper_authors=target_authors,
- author_cache=author_cache,
- quota_event=self.quota_exceeded_event,
- )
- if self.should_cancel:
- break
- if self.quota_exceeded_event.is_set():
- self._handle_quota_exceeded()
- return
- author_info_files.append(author_file)
+ if self.should_cancel:
+ break
+ if self.quota_exceeded_event.is_set():
+ self._handle_quota_exceeded()
+ return
+ author_info_files.append(author_file)
else:
self.log_manager.info("⏭ 跳过 Phase 2（skip_author_search=True）")
+
# Guard: 检查 Phase 1 是否爬取到任何引用文献
_total_citing = 0
for _cf, _ in citing_files:
@@ -878,12 +914,38 @@ def _fwd(p: Path) -> str:
"cost_summary": cost_summary,
}})
+ except asyncio.CancelledError:
+ self._save_resume_state(paper_groups, output_prefix, result_dir, total)
except Exception as e:
- self.log_manager.error(f"任务错误: {e}")
- import traceback; self.log_manager.error(traceback.format_exc())
- raise
+ if self.should_cancel:
+ self._save_resume_state(paper_groups, output_prefix, result_dir, total)
+ else:
+ self.log_manager.error(f"任务错误: {e}")
+ import traceback; self.log_manager.error(traceback.format_exc())
+ raise
finally:
self.is_running = False
+ self.should_cancel = False
+
+ def _save_resume_state(self, paper_groups, output_prefix, result_dir, total):
+ """取消后保存恢复状态（恢复时靠文件存在性自动跳过已完成阶段）"""
+ # 统计已完成的论文数（Phase 1 输出文件已存在）
+ done = sum(
+ 1 for i in range(total)
+ if (result_dir / f"paper{i+1}_citing.jsonl").exists()
+ )
+ self._resume_state = {
+ "paper_groups": paper_groups,
+ "output_prefix": output_prefix,
+ "result_dir": str(result_dir),
+ }
+ self.log_manager.warning(
+ f"⏸ 任务已取消（{done}/{total} 篇已完成），可点击「恢复」继续"
+ )
+ self.log_manager.broadcast_event("task_cancelled_resumable", {
+ "done": done,
+ "total": total,
+ })
def _handle_quota_exceeded(self):
"""Called when any phase signals that API quota is exhausted."""
@@ -972,6 +1034,12 @@ def cancel(self):
if self.is_running:
self.should_cancel = True
self.log_manager.warning("正在取消任务...")
+ # 强制取消底层 asyncio Task，防止阻塞在 await 上导致 is_running 永远为 True
+ if self.current_task and not self.current_task.done():
+ self.current_task.cancel()
+ else:
+ # 安全兜底：如果 is_running 已经是 False 但状态残留，重置
+ self.should_cancel = False
def get_status(self) -> dict:
"""
@@ -983,5 +1051,6 @@ def get_status(self) -> dict:
return {
"is_running": self.is_running,
"should_cancel": self.should_cancel,
- "has_stage1_result": self.stage1_result is not None
+ "has_stage1_result": self.stage1_result is not None,
+ "can_resume": self._resume_state is not None
}
diff --git a/citationclaw/static/js/main.js b/citationclaw/static/js/main.js
index 5f5c286..06ce051 100644
--- a/citationclaw/static/js/main.js
+++ b/citationclaw/static/js/main.js
@@ -385,6 +385,9 @@ async function resultsShowFolders() {
${folder.file_count} 个文件 · ${sizeMB} MB · ${date}
+ ${folder.has_meta ? `` : ''}
@@ -393,6 +396,36 @@ async function resultsShowFolders() {
const el = e.currentTarget;
resultsOpenFolder(el.dataset.folder, el.dataset.display);
});
+ const resumeFolderBtn = item.querySelector('[data-resume-folder]');
+ if (resumeFolderBtn) {
+ resumeFolderBtn.addEventListener('click', async (e) => {
+ e.stopPropagation();
+ const folderName = e.currentTarget.dataset.resumeFolder;
+ if (!confirm(`确定要从 "${folderName}" 恢复任务吗？已完成的阶段将自动跳过。`)) return;
+ try {
+ await saveIndexConfig();
+ const r = await fetch('/api/task/resume-from-folder', {
+ method: 'POST',
+ headers: { 'Content-Type': 'application/json' },
+ body: JSON.stringify({ folder: folderName })
+ });
+ const data = await r.json();
+ if (data.status === 'success') {
+ // 切换到 Home 面板显示运行状态
+ document.querySelector('[data-spa-panel="home"]').click();
+ const runBtn = document.getElementById('idx-run-btn');
+ runBtn.disabled = true;
+ runBtn.innerHTML = ' 运行中...';
+ document.getElementById('idx-cancel-btn').style.display = 'inline-flex';
+ document.getElementById('idx-resume-btn').style.display = 'none';
+ } else {
+ alert('恢复失败: ' + data.message);
+ }
+ } catch (err) {
+ alert('恢复失败：' + err.message);
+ }
+ });
+ }
item.querySelector('[data-delete]').addEventListener('click', async (e) => {
e.stopPropagation();
const name = e.currentTarget.dataset.delete;
@@ -550,6 +583,38 @@ function initIndexPage() {
};
let currentPhase = '处理中...';
+ // Search Model: select + 自定义输入联动
+ const _modelSelect = document.getElementById('idx-openai-model-select');
+ const _modelCustom = document.getElementById('idx-openai-model-custom');
+ const _modelHidden = document.getElementById('idx-openai-model');
+
+ function _syncModelHidden() {
+ if (_modelSelect.value === '__custom__') {
+ _modelCustom.style.display = '';
+ _modelHidden.value = _modelCustom.value;
+ } else {
+ _modelCustom.style.display = 'none';
+ _modelHidden.value = _modelSelect.value;
+ }
+ }
+ function _setModelFromValue(val) {
+ const opt = [..._modelSelect.options].find(o => o.value === val && o.value !== '__custom__');
+ if (opt) {
+ _modelSelect.value = val;
+ _modelCustom.style.display = 'none';
+ } else {
+ _modelSelect.value = '__custom__';
+ _modelCustom.style.display = '';
+ _modelCustom.value = val;
+ }
+ _modelHidden.value = val;
+ }
+ _modelSelect.addEventListener('change', () => {
+ if (_modelSelect.value === '__custom__') _modelCustom.focus();
+ _syncModelHidden();
+ });
+ _modelCustom.addEventListener('input', _syncModelHidden);
+
// 加载配置并填充 Home 面板表单
(async () => {
try {
@@ -560,7 +625,7 @@ function initIndexPage() {
el('idx-openai-key').value = cfg.openai_api_key || '';
_syncApiKeyType(el('idx-openai-key'));
el('idx-openai-url').value = cfg.openai_base_url || '';
- el('idx-openai-model').value = cfg.openai_model || '';
+ _setModelFromValue(cfg.openai_model || 'gemini-3-flash-preview-search');
el('idx-output-prefix').value = cfg.default_output_prefix || 'paper';
el('idx-renowned-scholar').checked = cfg.enable_renowned_scholar_filter !== false;
el('idx-author-verify').checked = cfg.enable_author_verification || false;
@@ -695,6 +760,10 @@ function initIndexPage() {
};
});
+ ws.on('task_cancelled_resumable', data => {
+ document.getElementById('idx-resume-btn').style.display = 'inline-flex';
+ });
+
ws.on('quota_exceeded', data => {
const msgEl = document.getElementById('quota-exceeded-message');
if (msgEl && data.message) {
@@ -748,6 +817,7 @@ function initIndexPage() {
runBtn.innerHTML = ' 运行中...';
document.getElementById('idx-cancel-btn').style.display = 'inline-flex';
+ document.getElementById('idx-resume-btn').style.display = 'none';
document.getElementById('idx-progress-section').style.display = 'block';
document.getElementById('idx-log-section').style.display = 'block';
document.getElementById('idx-results-section').style.display = 'none';
@@ -807,6 +877,30 @@ function initIndexPage() {
GlobalProgress.hide();
});
+ // 恢复按钮
+ document.getElementById('idx-resume-btn').addEventListener('click', async () => {
+ // 先保存当前 UI 配置（用户可能改了模型等参数）
+ await saveIndexConfig();
+ const resumeBtn = document.getElementById('idx-resume-btn');
+ resumeBtn.style.display = 'none';
+ runBtn.disabled = true;
+ runBtn.innerHTML = ' 运行中...';
+ document.getElementById('idx-cancel-btn').style.display = 'inline-flex';
+ startRunTimer();
+ try {
+ const resp = await fetch('/api/task/resume', { method: 'POST' });
+ const data = await resp.json();
+ if (data.status !== 'success') {
+ alert('恢复失败: ' + data.message);
+ resetRunBtn();
+ }
+ } catch (e) {
+ console.error('恢复失败:', e);
+ alert('恢复失败，请检查控制台');
+ resetRunBtn();
+ }
+ });
+
// 清空日志
document.getElementById('idx-clear-log-btn').addEventListener('click', () => {
document.getElementById('idx-log-container').innerHTML =
diff --git a/citationclaw/templates/index.html b/citationclaw/templates/index.html
index 9aa32c0..c436e81 100644
--- a/citationclaw/templates/index.html
+++ b/citationclaw/templates/index.html
@@ -156,13 +156,16 @@ 论文被引画像分析
-
@@ -210,6 +213,9 @@
论文被引画像分析
+