Skip to content

Commit 4a9643b

Browse files
author
Mingyue-Cheng
committed
Enhance CastClaw page with demo and paper showcase
1 parent 97bdbd8 commit 4a9643b

File tree

2 files changed

+124
-31
lines changed

2 files changed

+124
-31
lines changed

‎cast-claw/assets/castclaw-demo.mp4‎

31.8 MB
Binary file not shown.

‎cast-claw/index.html‎

Lines changed: 124 additions & 31 deletions
Original file line numberDiff line numberDiff line change
@@ -146,6 +146,59 @@
146146
font-size: 0.9rem; color: #555; line-height: 1.75; margin-bottom: 14px;
147147
}
148148
.example-card p:last-child { margin-bottom: 0; }
149+
.demo-video {
150+
width: 100%;
151+
display: block;
152+
margin-top: 14px;
153+
border-radius: 12px;
154+
border: 1px solid #e6e6e6;
155+
box-shadow: 0 6px 24px rgba(0,0,0,0.08);
156+
background: #000;
157+
}
158+
.paper-grid {
159+
display: grid;
160+
grid-template-columns: 1fr;
161+
gap: 18px;
162+
margin-top: 22px;
163+
}
164+
.paper-card {
165+
background: #fff;
166+
border: 1px solid #e6e6e6;
167+
border-radius: 14px;
168+
padding: 22px 22px 20px;
169+
box-shadow: 0 2px 10px rgba(0,0,0,0.05);
170+
border-top: 4px solid #d35400;
171+
}
172+
.paper-meta {
173+
display: inline-flex;
174+
align-items: center;
175+
border-radius: 999px;
176+
background: #fef3e2;
177+
color: #d35400;
178+
border: 1px solid #f5cba7;
179+
padding: 4px 10px;
180+
font-size: 0.76rem;
181+
font-weight: 700;
182+
margin-bottom: 12px;
183+
}
184+
.paper-title {
185+
font-size: 1rem;
186+
font-weight: 800;
187+
line-height: 1.55;
188+
color: #111;
189+
margin-bottom: 10px;
190+
}
191+
.paper-desc {
192+
font-size: 0.89rem;
193+
color: #555;
194+
line-height: 1.75;
195+
margin-bottom: 14px;
196+
}
197+
.paper-links {
198+
display: flex;
199+
flex-wrap: wrap;
200+
gap: 10px;
201+
}
149202

150203
/* ── Workflow stepper ───────────────────────── */
151204
.workflow { margin-top: 28px; display: flex; flex-direction: column; gap: 0; }
@@ -183,11 +236,6 @@
183236
}
184237
.agent-card .agent-icon { font-size: 2rem; margin-bottom: 10px; }
185238
.agent-card h4 { font-size: 1rem; font-weight: 800; color: #111; margin-bottom: 4px; }
186-
.agent-card .agent-shortcut {
187-
font-size: 0.78rem; color: #d35400; font-weight: 600;
188-
background: #fef3e2; border: 1px solid #f5cba7; border-radius: 4px;
189-
padding: 2px 8px; display: inline-block; margin-bottom: 10px;
190-
}
191239
.agent-card p { font-size: 0.88rem; color: #555; line-height: 1.65; margin: 0; }
192240

193241
/* ── Highlight panels ───────────────────────── */
@@ -332,7 +380,7 @@
332380
@media (max-width: 640px) {
333381
header h1 { font-size: 1.5rem; }
334382
section { padding: 48px 0; }
335-
.feature-grid, .agent-grid { grid-template-columns: 1fr; }
383+
.feature-grid, .agent-grid, .paper-grid { grid-template-columns: 1fr; }
336384
.highlight-panel { padding: 20px 18px 22px; }
337385
.toolbox-group { padding: 18px 18px 20px; }
338386
.toolbox-card { padding: 18px 18px 20px; }
@@ -430,21 +478,18 @@ <h3 class="highlight-title">多智能体协同</h3>
430478
<div class="agent-card">
431479
<div class="agent-icon">🗂️</div>
432480
<h4>Planner（规划者）</h4>
433-
<span class="agent-shortcut">Ctrl+1</span>
434481
<p>负责任务定义、数据诊断与阶段编排。并发启动两条分析轨道：定性域研究（WebSearch）+ 定量数据统计，融合为预测前报告。生成 2–4 个技能文件供人工审核确认。</p>
435482
</div>
436483

437484
<div class="agent-card">
438485
<div class="agent-icon">🔬</div>
439486
<h4>Forecaster（预测者）</h4>
440-
<span class="agent-shortcut">Ctrl+2</span>
441487
<p>驱动迭代实验循环——读取最佳结果与失败历史，从技能中选取模型配置，调用 <code>generate_model</code> 训练评估，进行反思记录，管理实验预算。停滞时触发 HITL 暂停等待人类反馈。</p>
442488
</div>
443489

444490
<div class="agent-card">
445491
<div class="agent-icon">📊</div>
446492
<h4>Critic（评审者）</h4>
447-
<span class="agent-shortcut">Ctrl+3</span>
448493
<p>读取所有实验产物，生成各模型族最佳结果对比、按时序特征的性能分解、可视化脚本，以及结构化的最终 Markdown 预测报告，输出至 <code>.forecast/reports/final-report.md</code>。</p>
449494
</div>
450495

@@ -648,7 +693,7 @@ <h4>崩溃隔离</h4>
648693
<div class="toolbox-group toolbox-group-analytics">
649694
<div class="toolbox-group-head">
650695
<div class="toolbox-group-index">B</div>
651-
<h3 class="toolbox-group-title">分析与模型类</h3>
696+
<h3 class="toolbox-group-title">预测辅助工具</h3>
652697
</div>
653698
<p class="toolbox-group-desc">将数据分析、时序特征诊断与模型资产封装成可组合的能力层，支撑技能生成、路线筛选与不同模型族的系统探索。</p>
654699

@@ -704,26 +749,12 @@ <h4>统计模型</h4>
704749
<div class="model-group">
705750
<h4>深度学习模型</h4>
706751
<div class="model-pills">
707-
<span class="model-pill">DLinear</span>
708-
<span class="model-pill">NLinear</span>
709-
<span class="model-pill">PatchTST</span>
710-
<span class="model-pill">TimesNet</span>
711-
<span class="model-pill">iTransformer</span>
712-
<span class="model-pill">Autoformer</span>
713-
<span class="model-pill">FEDformer</span>
714752
<span class="model-pill">Informer</span>
715-
<span class="model-pill">Transformer</span>
716-
<span class="model-pill">MICN</span>
717-
<span class="model-pill">SCINet</span>
718-
<span class="model-pill">TimeMixer</span>
719-
<span class="model-pill">TSMixer</span>
720-
<span class="model-pill">SegRNN</span>
721-
<span class="model-pill">FITS</span>
753+
<span class="model-pill">PatchTST</span>
754+
<span class="model-pill">TSMixer++</span>
722755
<span class="model-pill">TiDE</span>
723-
<span class="model-pill">Mamba</span>
724-
<span class="model-pill">Crossformer</span>
725756
<span class="model-pill">ConvTimeNet</span>
726-
<span class="model-pill">20+ 更多</span>
757+
<span class="model-pill">等</span>
727758
</div>
728759
</div>
729760

@@ -940,6 +971,18 @@ <h3 class="sub-title">样例数据集</h3>
940971
</div>
941972
</div>
942973

974+
<h3 class="sub-title">演示视频</h3>
975+
<div class="example-showcase">
976+
<div class="example-card">
977+
<span class="example-badge">Product Demo</span>
978+
<p>下面的视频展示了 CastClaw 在真实终端工作流中的使用方式，包括任务建立、智能体协同切换，以及预测流程中的关键交互环节。你也可以直接下载原始压缩版演示文件 <a href="assets/castclaw-demo.mp4" download><code>castclaw-demo.mp4</code></a>。</p>
979+
<video class="demo-video" controls preload="metadata" playsinline>
980+
<source src="assets/castclaw-demo.mp4" type="video/mp4" />
981+
你的浏览器暂不支持视频播放，请直接下载演示文件查看。
982+
</video>
983+
</div>
984+
</div>
985+
943986
<h3 class="sub-title">安装</h3>
944987
<div class="code-block"><span class="comment"># 方式一：npm 全局安装（推荐）</span>
945988
<span class="cmd">npm</span> install -g castclaw
@@ -1038,11 +1081,61 @@ <h3 class="sub-title">使用样例展示</h3>
10381081
<div class="container">
10391082
<h2 class="section-title">相关项目</h2>
10401083
<p>
1041-
CastClaw 与 <strong>AlphaCast</strong>、<strong>Cast-R1</strong>、<strong>Agent-R1</strong> 共同构成我们在时间序列智能体与相关训练方法方向上的研究工作，分别从人机协同实验编排、预测系统探索、顺序决策学习与智能体训练框架等不同侧面推进这一研究方向。
1084+
CastClaw 所在的研究线并不局限于单一项目，而是与一系列围绕时间序列智能体、交互式预测、推理增强和经验进化展开的代表性论文共同构成。下面列出几篇与 CastClaw 关系最紧密的工作，更多论文可以在 <a href="https://mingyue-cheng.github.io/" target="_blank" rel="noopener">Mingyue Cheng 的主页</a> 中继续检索。
10421085
</p>
1043-
<div class="pill-links">
1044-
<span class="pill pill-out">AlphaCast</span>
1045-
<a class="pill pill-out" href="../agent-apps/#cast-r1">Cast-R1</a>
1086+
1087+
<div class="paper-grid">
1088+
<div class="paper-card">
1089+
<span class="paper-meta">Position Paper</span>
1090+
<h3 class="paper-title">Beyond Model-Centric Prediction — Agentic Time Series Forecasting</h3>
1091+
<p class="paper-desc">从范式层面提出 Agentic Time Series Forecasting，为 CastClaw 这种“预测即交互式决策过程”的设计提供理论起点。</p>
1092+
<div class="paper-links">
1093+
<a class="pill pill-orange" href="https://arxiv.org/abs/2602.01776" target="_blank" rel="noopener">PDF</a>
1094+
</div>
1095+
</div>
1096+
1097+
<div class="paper-card">
1098+
<span class="paper-meta">Human-AI Collaboration</span>
1099+
<h3 class="paper-title">AlphaCast: A Human Wisdom-LLM Intelligence Co-Reasoning Framework for Interactive Time Series Forecasting</h3>
1100+
<p class="paper-desc">聚焦人机协同交互式预测，强调在关键节点引入专家判断，是 CastClaw 的人机协作设计来源之一。</p>
1101+
<div class="paper-links">
1102+
<a class="pill pill-orange" href="https://arxiv.org/abs/2511.08947" target="_blank" rel="noopener">PDF</a>
1103+
<a class="pill pill-out" href="https://github.com/SkyeGT/AlphaCast_Official" target="_blank" rel="noopener">Code</a>
1104+
</div>
1105+
</div>
1106+
1107+
<div class="paper-card">
1108+
<span class="paper-meta">Sequential Decision Policy</span>
1109+
<h3 class="paper-title">Cast-R1: Learning Tool-Augmented Sequential Decision Policies for Time Series Forecasting</h3>
1110+
<p class="paper-desc">从强化学习与工具调用角度刻画时序预测中的多步决策，为 CastClaw 的实验循环和策略优化提供方法支撑。</p>
1111+
<div class="paper-links">
1112+
<a class="pill pill-orange" href="https://arxiv.org/abs/2602.13802" target="_blank" rel="noopener">PDF</a>
1113+
<a class="pill pill-out" href="https://github.com/Xiaoyu-Tao/Cast-R1-TS" target="_blank" rel="noopener">Code</a>
1114+
</div>
1115+
</div>
1116+
1117+
<div class="paper-card">
1118+
<span class="paper-meta">Experience & Memory</span>
1119+
<h3 class="paper-title">MemCast: Memory-Driven Time Series Forecasting with Experience-Conditioned Reasoning</h3>
1120+
<p class="paper-desc">强调经验积累、记忆调用与持续进化，对 CastClaw 中基于 Skill 的长期能力沉淀具有直接启发意义。</p>
1121+
<div class="paper-links">
1122+
<a class="pill pill-orange" href="https://arxiv.org/abs/2602.03164" target="_blank" rel="noopener">PDF</a>
1123+
<a class="pill pill-out" href="https://github.com/Xiaoyu-Tao/MemCast-TS" target="_blank" rel="noopener">Code</a>
1124+
</div>
1125+
</div>
1126+
1127+
<div class="paper-card">
1128+
<span class="paper-meta">Reasoning Over Time</span>
1129+
<h3 class="paper-title">Can Slow-Thinking LLMs Reason Over Time? Empirical Studies in Time Series Forecasting</h3>
1130+
<p class="paper-desc">系统研究“慢思考”式推理在时间序列预测中的作用，为 CastClaw 的分析、反思与多轮决策机制提供经验依据。</p>
1131+
<div class="paper-links">
1132+
<a class="pill pill-orange" href="https://arxiv.org/abs/2505.24511" target="_blank" rel="noopener">PDF</a>
1133+
</div>
1134+
</div>
1135+
</div>
1136+
1137+
<div class="pill-links" style="margin-top:22px;">
1138+
<a class="pill pill-out" href="https://mingyue-cheng.github.io/" target="_blank" rel="noopener">更多论文</a>
10461139
<a class="pill pill-out" href="../agent-r1/">Agent-R1</a>
10471140
<a class="pill pill-out" href="../">← 返回主页</a>
10481141
</div>

0 commit comments

Comments
 (0)