Add daily arXiv digest pipeline and heartbeat push workflow

HEARTBEAT.md

@@ -23,3 +23,22 @@
 
 - 需要主动触达时：直接给用户发送自然消息，不要输出 HEARTBEAT_OK。
 - 若本轮不需要触达：回复 HEARTBEAT_OK。
+
+---
+
+## 每日 ArXiv 简报任务（新增）
+
+目标：每天自动产出并推送一条“最新+热度”论文简报给谷老板。
+
+执行规则：
+1. 每天 09:30 之后（且 23:00 前），若今日尚未推送，则运行：
+   - `python3 /Users/guchen/.openclaw/workspace/org/cases/arxiv_digest/arxiv_daily_digest.py`
+2. 读取生成文件：
+   - `/Users/guchen/.openclaw/workspace/org/cases/arxiv_digest/output/YYYY-MM-DD.md`
+3. 用聊天口吻给谷老板推送：
+   - 今日 Top3 热点论文（标题+一句话）
+   - 1 条 Val 建议
+4. 推送完成后写入状态文件：
+   - `/Users/guchen/.openclaw/workspace/org/cases/arxiv_digest/state/last_push.json`
+   - 内容至少包含 `{ "date": "YYYY-MM-DD", "pushed": true }`
+5. 同一天不重复推送；若脚本失败，发一条简短异常提示并稍后重试。

org/cases/arxiv_digest/README.md

@@ -0,0 +1,19 @@
+# ArXiv Daily Digest (Val)
+
+## What it does
+- Fetches latest arXiv papers for: cs.AI/cs.LG/cs.CL/cs.CV/cs.RO/stat.ML
+- Ranks "hot" papers with a practical score (recency + keyword signal + HN mention proxy + code hint)
+- Produces daily markdown brief
+
+## Run
+```bash
+python3 /Users/guchen/.openclaw/workspace/org/cases/arxiv_digest/arxiv_daily_digest.py
+```
+
+## Output
+- Daily brief: `output/YYYY-MM-DD.md`
+- Latest mirror: `output/latest.md`
+- Runtime state: `state/latest_state.json`
+
+## Notes
+- Hotness is heuristic (v1). Can be upgraded with stronger social/citation signals later.

org/cases/arxiv_digest/arxiv_daily_digest.py

@@ -0,0 +1,168 @@
+#!/usr/bin/env python3
+import datetime as dt
+import json
+import re
+import urllib.parse
+import urllib.request
+import xml.etree.ElementTree as ET
+from pathlib import Path
+
+BASE = Path('/Users/guchen/.openclaw/workspace/org/cases/arxiv_digest')
+OUT_DIR = BASE / 'output'
+STATE_DIR = BASE / 'state'
+OUT_DIR.mkdir(parents=True, exist_ok=True)
+STATE_DIR.mkdir(parents=True, exist_ok=True)
+
+CATS = ['cs.AI', 'cs.LG', 'cs.CL', 'cs.CV', 'cs.RO', 'stat.ML']
+MAX_RESULTS = 120
+HOT_KEYWORDS = ['agent', 'reasoning', 'multimodal', 'alignment', 'rl', 'diffusion', 'transformer', 'benchmark']
+
+
+def fetch(url, timeout=20):
+    req = urllib.request.Request(url, headers={'User-Agent': 'Val-ArXiv-Digest/1.0'})
+    with urllib.request.urlopen(req, timeout=timeout) as r:
+        return r.read().decode('utf-8', errors='ignore')
+
+
+def query_arxiv():
+    q = ' OR '.join([f'cat:{c}' for c in CATS])
+    params = {
+        'search_query': q,
+        'start': 0,
+        'max_results': MAX_RESULTS,
+        'sortBy': 'submittedDate',
+        'sortOrder': 'descending',
+    }
+    url = 'http://export.arxiv.org/api/query?' + urllib.parse.urlencode(params)
+    xml = fetch(url)
+    ns = {'a': 'http://www.w3.org/2005/Atom', 'arxiv': 'http://arxiv.org/schemas/atom'}
+    root = ET.fromstring(xml)
+    papers = []
+    for e in root.findall('a:entry', ns):
+        pid = (e.findtext('a:id', default='', namespaces=ns) or '').strip()
+        title = re.sub(r'\s+', ' ', (e.findtext('a:title', default='', namespaces=ns) or '').strip())
+        summary = re.sub(r'\s+', ' ', (e.findtext('a:summary', default='', namespaces=ns) or '').strip())
+        published = e.findtext('a:published', default='', namespaces=ns)
+        updated = e.findtext('a:updated', default='', namespaces=ns)
+        authors = [re.sub(r'\s+', ' ', (a.findtext('a:name', default='', namespaces=ns) or '').strip())
+                   for a in e.findall('a:author', ns)]
+        primary = ''
+        pcat = e.find('arxiv:primary_category', ns)
+        if pcat is not None:
+            primary = pcat.attrib.get('term', '')
+        all_cats = [c.attrib.get('term', '') for c in e.findall('a:category', ns)]
+        papers.append({
+            'id': pid,
+            'arxiv_id': pid.split('/abs/')[-1] if '/abs/' in pid else pid,
+            'title': title,
+            'summary': summary,
+            'published': published,
+            'updated': updated,
+            'authors': authors,
+            'primary': primary,
+            'categories': all_cats,
+        })
+    return papers
+
+
+def hours_since(iso):
+    try:
+        t = dt.datetime.fromisoformat(iso.replace('Z', '+00:00'))
+        now = dt.datetime.now(dt.timezone.utc)
+        return max(0.0, (now - t).total_seconds() / 3600)
+    except Exception:
+        return 999.0
+
+
+def hn_hits_for(paper):
+    # lightweight buzz proxy
+    q = paper['arxiv_id']
+    url = 'https://hn.algolia.com/api/v1/search?' + urllib.parse.urlencode({'query': q, 'tags': 'story'})
+    try:
+        data = json.loads(fetch(url, timeout=3))
+        return int(data.get('nbHits', 0))
+    except Exception:
+        return 0
+
+
+def score(p):
+    h = hours_since(p['published'])
+    recency = max(0, 72 - h) / 72 * 60
+    kw = sum(1 for k in HOT_KEYWORDS if k in (p['title'] + ' ' + p['summary']).lower())
+    keyword_score = min(20, kw * 4)
+    github_bonus = 8 if ('github.com' in p['summary'].lower() or 'code:' in p['summary'].lower()) else 0
+    hn = hn_hits_for(p)
+    hn_score = min(12, hn * 2)
+    cat_bonus = 5 if p['primary'] in ['cs.AI', 'cs.LG', 'cs.CL'] else 2
+    total = recency + keyword_score + github_bonus + hn_score + cat_bonus
+    return round(total, 2), {'recency': round(recency,2), 'keywords': keyword_score, 'github': github_bonus, 'hn': hn_score, 'cat': cat_bonus}
+
+
+def one_liner(p):
+    t = p['title']
+    if len(t) > 90:
+        t = t[:87] + '...'
+    return f"{t}（{p['primary']}）"
+
+
+def build_digest(papers):
+    # limit scoring set for speed/stability
+    recent_pool = sorted(papers, key=lambda x: x['published'], reverse=True)[:40]
+    scored = []
+    for p in recent_pool:
+        s, detail = score(p)
+        p2 = dict(p)
+        p2['hot_score'] = s
+        p2['score_detail'] = detail
+        scored.append(p2)
+    hot = sorted(scored, key=lambda x: x['hot_score'], reverse=True)[:5]
+    latest = sorted(scored, key=lambda x: x['published'], reverse=True)[:10]
+    return hot, latest, scored
+
+
+def to_md(hot, latest):
+    today = dt.datetime.now().strftime('%Y-%m-%d')
+    lines = []
+    lines.append(f"# ArXiv Daily Brief - {today}")
+    lines.append('')
+    lines.append('## 🔥 今日热度 Top 5（新鲜度+关键词+HN提及+代码线索）')
+    for i, p in enumerate(hot, 1):
+        lines.append(f"{i}. **{p['title']}**")
+        lines.append(f"   - arXiv: {p['id']}")
+        lines.append(f"   - 类别: {p['primary']} | HotScore: {p['hot_score']} | 作者: {', '.join(p['authors'][:3])}")
+        lines.append(f"   - 速读: {one_liner(p)}")
+    lines.append('')
+    lines.append('## 🆕 最新上新 Top 10')
+    for i, p in enumerate(latest, 1):
+        lines.append(f"{i}. {p['title']} ({p['primary']}) - {p['id']}")
+    lines.append('')
+    lines.append('## Val 今日建议')
+    lines.append('- 先读 Top 5 里的 1-2 篇，优先看是否有可直接复用的方法/代码。')
+    lines.append('- 若你愿意，我下一步可对 Top 3 产出“中文三段式精读卡”（问题-方法-可落地点）。')
+    return '\n'.join(lines) + '\n'
+
+
+def main():
+    papers = query_arxiv()
+    hot, latest, scored = build_digest(papers)
+    md = to_md(hot, latest)
+
+    day = dt.datetime.now().strftime('%Y-%m-%d')
+    out = OUT_DIR / f'{day}.md'
+    out.write_text(md, encoding='utf-8')
+    (OUT_DIR / 'latest.md').write_text(md, encoding='utf-8')
+
+    state = {
+        'updatedAt': dt.datetime.now().isoformat(),
+        'date': day,
+        'papersFetched': len(papers),
+        'topHot': [
+            {'title': p['title'], 'arxiv_id': p['arxiv_id'], 'score': p['hot_score']} for p in hot
+        ]
+    }
+    (STATE_DIR / 'latest_state.json').write_text(json.dumps(state, ensure_ascii=False, indent=2), encoding='utf-8')
+    print(str(out))
+
+
+if __name__ == '__main__':
+    main()

org/cases/arxiv_digest/output/2026-03-08.md

@@ -0,0 +1,39 @@
+# ArXiv Daily Brief - 2026-03-08
+
+## 🔥 今日热度 Top 5（新鲜度+关键词+HN提及+代码线索）
+1. **SurvHTE-Bench: A Benchmark for Heterogeneous Treatment Effect Estimation in Survival Analysis**
+   - arXiv: http://arxiv.org/abs/2603.05483v1
+   - 类别: cs.LG | HotScore: 33.06 | 作者: Shahriar Noroozizadeh, Xiaobin Shen, Jeremy C. Weiss
+   - 速读: SurvHTE-Bench: A Benchmark for Heterogeneous Treatment Effect Estimation in Survival An...（cs.LG）
+2. **Accelerating Text-to-Video Generation with Calibrated Sparse Attention**
+   - arXiv: http://arxiv.org/abs/2603.05503v1
+   - 类别: cs.CV | HotScore: 28.17 | 作者: Shai Yehezkel, Shahar Yadin, Noam Elata
+   - 速读: Accelerating Text-to-Video Generation with Calibrated Sparse Attention（cs.CV）
+3. **Observing and Controlling Features in Vision-Language-Action Models**
+   - arXiv: http://arxiv.org/abs/2603.05487v1
+   - 类别: cs.RO | HotScore: 28.08 | 作者: Hugo Buurmeijer, Carmen Amo Alonso, Aiden Swann
+   - 速读: Observing and Controlling Features in Vision-Language-Action Models（cs.RO）
+4. **Towards Provably Unbiased LLM Judges via Bias-Bounded Evaluation**
+   - arXiv: http://arxiv.org/abs/2603.05485v1
+   - 类别: cs.AI | HotScore: 27.06 | 作者: Benjamin Feuer, Lucas Rosenblatt, Oussama Elachqar
+   - 速读: Towards Provably Unbiased LLM Judges via Bias-Bounded Evaluation（cs.AI）
+5. **An interpretable prototype parts-based neural network for medical tabular data**
+   - arXiv: http://arxiv.org/abs/2603.05423v1
+   - 类别: cs.LG | HotScore: 26.1 | 作者: Jacek Karolczak, Jerzy Stefanowski
+   - 速读: An interpretable prototype parts-based neural network for medical tabular data（cs.LG）
+
+## 🆕 最新上新 Top 10
+1. Transformer-Based Inpainting for Real-Time 3D Streaming in Sparse Multi-Camera Setups (cs.CV) - http://arxiv.org/abs/2603.05507v1
+2. FaceCam: Portrait Video Camera Control via Scale-Aware Conditioning (cs.CV) - http://arxiv.org/abs/2603.05506v1
+3. RoboPocket: Improve Robot Policies Instantly with Your Phone (cs.RO) - http://arxiv.org/abs/2603.05504v1
+4. Accelerating Text-to-Video Generation with Calibrated Sparse Attention (cs.CV) - http://arxiv.org/abs/2603.05503v1
+5. POET-X: Memory-efficient LLM Training by Scaling Orthogonal Transformation (cs.LG) - http://arxiv.org/abs/2603.05500v1
+6. The Spike, the Sparse and the Sink: Anatomy of Massive Activations and Attention Sinks (cs.AI) - http://arxiv.org/abs/2603.05498v1
+7. Safe-SAGE: Social-Semantic Adaptive Guidance for Safe Engagement through Laplace-Modulated Poisson Safety Functions (cs.RO) - http://arxiv.org/abs/2603.05497v1
+8. Cheap Thrills: Effective Amortized Optimization Using Inexpensive Labels (cs.LG) - http://arxiv.org/abs/2603.05495v1
+9. Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation (cs.LG) - http://arxiv.org/abs/2603.05494v1
+10. cuRoboV2: Dynamics-Aware Motion Generation with Depth-Fused Distance Fields for High-DoF Robots (cs.RO) - http://arxiv.org/abs/2603.05493v1
+
+## Val 今日建议
+- 先读 Top 5 里的 1-2 篇，优先看是否有可直接复用的方法/代码。
+- 若你愿意，我下一步可对 Top 3 产出“中文三段式精读卡”（问题-方法-可落地点）。

org/cases/arxiv_digest/output/latest.md

@@ -0,0 +1,39 @@
+# ArXiv Daily Brief - 2026-03-08
+
+## 🔥 今日热度 Top 5（新鲜度+关键词+HN提及+代码线索）
+1. **SurvHTE-Bench: A Benchmark for Heterogeneous Treatment Effect Estimation in Survival Analysis**
+   - arXiv: http://arxiv.org/abs/2603.05483v1
+   - 类别: cs.LG | HotScore: 33.06 | 作者: Shahriar Noroozizadeh, Xiaobin Shen, Jeremy C. Weiss
+   - 速读: SurvHTE-Bench: A Benchmark for Heterogeneous Treatment Effect Estimation in Survival An...（cs.LG）
+2. **Accelerating Text-to-Video Generation with Calibrated Sparse Attention**
+   - arXiv: http://arxiv.org/abs/2603.05503v1
+   - 类别: cs.CV | HotScore: 28.17 | 作者: Shai Yehezkel, Shahar Yadin, Noam Elata
+   - 速读: Accelerating Text-to-Video Generation with Calibrated Sparse Attention（cs.CV）
+3. **Observing and Controlling Features in Vision-Language-Action Models**
+   - arXiv: http://arxiv.org/abs/2603.05487v1
+   - 类别: cs.RO | HotScore: 28.08 | 作者: Hugo Buurmeijer, Carmen Amo Alonso, Aiden Swann
+   - 速读: Observing and Controlling Features in Vision-Language-Action Models（cs.RO）
+4. **Towards Provably Unbiased LLM Judges via Bias-Bounded Evaluation**
+   - arXiv: http://arxiv.org/abs/2603.05485v1
+   - 类别: cs.AI | HotScore: 27.06 | 作者: Benjamin Feuer, Lucas Rosenblatt, Oussama Elachqar
+   - 速读: Towards Provably Unbiased LLM Judges via Bias-Bounded Evaluation（cs.AI）
+5. **An interpretable prototype parts-based neural network for medical tabular data**
+   - arXiv: http://arxiv.org/abs/2603.05423v1
+   - 类别: cs.LG | HotScore: 26.1 | 作者: Jacek Karolczak, Jerzy Stefanowski
+   - 速读: An interpretable prototype parts-based neural network for medical tabular data（cs.LG）
+
+## 🆕 最新上新 Top 10
+1. Transformer-Based Inpainting for Real-Time 3D Streaming in Sparse Multi-Camera Setups (cs.CV) - http://arxiv.org/abs/2603.05507v1
+2. FaceCam: Portrait Video Camera Control via Scale-Aware Conditioning (cs.CV) - http://arxiv.org/abs/2603.05506v1
+3. RoboPocket: Improve Robot Policies Instantly with Your Phone (cs.RO) - http://arxiv.org/abs/2603.05504v1
+4. Accelerating Text-to-Video Generation with Calibrated Sparse Attention (cs.CV) - http://arxiv.org/abs/2603.05503v1
+5. POET-X: Memory-efficient LLM Training by Scaling Orthogonal Transformation (cs.LG) - http://arxiv.org/abs/2603.05500v1
+6. The Spike, the Sparse and the Sink: Anatomy of Massive Activations and Attention Sinks (cs.AI) - http://arxiv.org/abs/2603.05498v1
+7. Safe-SAGE: Social-Semantic Adaptive Guidance for Safe Engagement through Laplace-Modulated Poisson Safety Functions (cs.RO) - http://arxiv.org/abs/2603.05497v1
+8. Cheap Thrills: Effective Amortized Optimization Using Inexpensive Labels (cs.LG) - http://arxiv.org/abs/2603.05495v1
+9. Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation (cs.LG) - http://arxiv.org/abs/2603.05494v1
+10. cuRoboV2: Dynamics-Aware Motion Generation with Depth-Fused Distance Fields for High-DoF Robots (cs.RO) - http://arxiv.org/abs/2603.05493v1
+
+## Val 今日建议
+- 先读 Top 5 里的 1-2 篇，优先看是否有可直接复用的方法/代码。
+- 若你愿意，我下一步可对 Top 3 产出“中文三段式精读卡”（问题-方法-可落地点）。

org/cases/arxiv_digest/state/latest_state.json

@@ -0,0 +1,32 @@
+{
+  "updatedAt": "2026-03-08T14:48:22.617040",
+  "date": "2026-03-08",
+  "papersFetched": 120,
+  "topHot": [
+    {
+      "title": "SurvHTE-Bench: A Benchmark for Heterogeneous Treatment Effect Estimation in Survival Analysis",
+      "arxiv_id": "2603.05483v1",
+      "score": 33.06
+    },
+    {
+      "title": "Accelerating Text-to-Video Generation with Calibrated Sparse Attention",
+      "arxiv_id": "2603.05503v1",
+      "score": 28.17
+    },
+    {
+      "title": "Observing and Controlling Features in Vision-Language-Action Models",
+      "arxiv_id": "2603.05487v1",
+      "score": 28.08
+    },
+    {
+      "title": "Towards Provably Unbiased LLM Judges via Bias-Bounded Evaluation",
+      "arxiv_id": "2603.05485v1",
+      "score": 27.06
+    },
+    {
+      "title": "An interpretable prototype parts-based neural network for medical tabular data",
+      "arxiv_id": "2603.05423v1",
+      "score": 26.1
+    }
+  ]
+}