Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

speech-paper-daily

v1.0.0

语音领域每日论文速递。搜索最新语音大模型(Speech LLM、TTS、ASR、codec、speech generation)和语音前端(speech enhancement、noise suppression、beamforming、source separation、dereverberation)预印本论...

1· 97·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for jusperlee/speech-paper-daily.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "speech-paper-daily" (jusperlee/speech-paper-daily) from ClawHub.
Skill page: https://clawhub.ai/jusperlee/speech-paper-daily
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install speech-paper-daily

ClawHub CLI

Package manager switcher

npx clawhub@latest install speech-paper-daily
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
Name/description match the actions in SKILL.md: the skill scrapes arXiv, reads papers, summarizes them, and writes a document to a specified Tencent Docs folder. No unrelated binaries, env vars, or installs are requested.
Instruction Scope
Instructions are focused on the stated task (fetch arXiv / read papers / extract links / produce expert summaries / write to Tencent Docs). They require using platform tools (web_fetch, read_arxiv_paper, search_arxiv, write, exec) and instruct writing temporary files under /tmp and executing a generated Python script which calls 'mcporter' to create the Tencent Docs entry. This is expected for the write-to-docs step, but it does mean the agent will create and execute code at runtime and invoke an external docs API — verify you trust the platform tools (mcporter/tencent-docs) that handle credentials.
Install Mechanism
Instruction-only skill with no install spec and no code files; nothing is downloaded or installed by the skill itself.
Credentials
No environment variables, credentials, or config paths are declared. The instructions rely on the platform's existing tooling (mcporter) to access Tencent Docs; the absence of declared credentials is reasonable for an instruction-only skill, but you should confirm how your platform provides and scopes the docs integration credentials.
Persistence & Privilege
Skill is not always-enabled and does not request persistent system-wide changes. It writes only to temporary files under /tmp during operation and invokes platform tools to publish documents to a specific folder ID.
Assessment
This skill appears coherent for collecting and summarizing arXiv speech papers and posting them into a Tencent Docs folder. Before installing: (1) confirm you trust the platform tools (mcporter / tencent-docs) since the skill will invoke them to create documents in folder ID 'YUsookchBhki'; (2) understand the skill will write temporary files to /tmp and execute a generated Python script (ensure execution is permitted and safe in your environment); (3) if you don't want automatic writes to your real folder, change the folder ID to a test folder or run the skill in a sandbox first; (4) verify the platform's credentials for Tencent Docs are scoped minimally (only permission to write to the intended folder). If any of those are unacceptable, do not enable the skill or run it only in a reviewed/test mode.

Like a lobster shell, security has layers — review code before you run it.

latestvk9748dz0v0q2psrq0gh55cd0hh83fkqk
97downloads
1stars
1versions
Updated 1mo ago
v1.0.0
MIT-0

语音论文速递 Skill

目标

搜索近 30 天语音领域 arXiv 预印本(20-30 篇),以领域专家视角精读,写入腾讯文档。


第一步:获取论文列表

主要来源(优先使用):用 web_fetch 抓取 arXiv 官方每日列表页面,获取当天最新论文 ID:

  1. https://arxiv.org/list/cs.SD/new — Sound 分类(仅今日新提交)
  2. https://arxiv.org/list/eess.AS/new — Audio and Speech Processing 分类(仅今日新提交)

从页面中提取当天所有 arXiv ID,合并去重。/new 页面只列出今日新提交论文,无需按日期筛选。

⚠️ 页面只显示 ID,不含 abstract。提取 ID 后,用 read_arxiv_paper 逐篇读取全文(含 abstract + 正文)。

补充来源(当天论文 < 5 篇时启用):使用 search_arxivdate_from 往前推 7 天,关键词:

  • speech synthesis TTS neural
  • automatic speech recognition ASR
  • speech enhancement noise suppression
  • speech separation audio

⚠️ 过滤规则(必须执行)

从官方列表获取的论文已属于 cs.SDeess.AS,无需额外过滤分类。但需人工判断是否与语音/音频处理直接相关,丢弃以下类型:

  • 纯音乐生成(与语音无关)
  • 纯图像/视频处理(误入 cross-list)
  • 纯理论数学/物理声学(非 ML/DL 方法)

保留所有 TTS、ASR、语音增强、语音分离、说话人识别/验证、音频语言模型、声码器、语音编解码等方向的论文。

合并两个页面结果,按提交日期降序,去重后全部保留(不设数量上限)。


第二步:精读论文

对所有通过过滤的论文,无论评分高低,一律读取全文,策略如下:

  1. 优先用 HTML 版本:用 web_fetch 抓取 https://arxiv.org/html/<ID>v1,文字质量更好、公式不乱码
  2. 回退 PDF:若 HTML 页面返回 404 或内容为空,改用 read_arxiv_paper 读 PDF

读取全文时,注意提取论文中出现的 demo 页面链接(通常在 abstract 或 introduction 中,形如 demo pageaudio sampleshttps://xxx.github.io 等)和代码仓库链接(形如 github.com/xxx)。

语音领域专家视角输出:

## [序号] 论文标题

**arXiv ID**: 2501.xxxxx
**方向**: 语音大模型 / 语音前端
**作者**: 作者1, 作者2, 作者3 等
**机构**: xxx(作者所属单位,多机构用 / 分隔)
**发布日期**: YYYY-MM-DD
**论文链接**: https://arxiv.org/abs/2501.xxxxx
**PDF 链接**: https://arxiv.org/pdf/2501.xxxxx.pdf
**代码链接**: https://github.com/xxx/xxx(若论文未提供则填"暂无")
**Demo 链接**: https://xxx.github.io/xxx(若论文未提供则填"暂无")

### 📌 简介
2-3句话:解决什么问题,核心贡献是什么。

### 🔧 技术方案

**模型架构**:
- 整体框架(encoder/decoder结构、主干网络类型)
- 关键模块设计(注意力机制、特征提取方式、信号处理流程)
- 与 Transformer/Conformer/Mamba 等基础架构的关系

**核心创新**:
- 本文提出的新方法/新机制(与现有方法的本质区别)
- 解决了什么已有方法解决不了的问题

**训练策略**:
- 损失函数设计(感知损失、对抗损失、重建损失等)
- 数据预处理/增强方式
- 预训练 / 微调 / 多阶段训练策略(如有)

### 📊 实验结果
- 数据集 + 主要指标数值(与 baseline 对比)
- 是否开源

### ⭐ 评分:X/10
理由:创新性 / 实验充分性 / 实用价值

评分标准

分数标准
9-10突破性,顶会水准(Interspeech/ICASSP/NeurIPS)
7-8有实质贡献,实验较充分
5-6增量工作,有参考价值
3-4实验不足或方法普通
1-2质量较低,建议跳过

第三步:写入腾讯文档

固定参数(勿改)

  • 文件夹 IDYUsookchBhki(「每日论文速递」文件夹,已确认)
  • 文档标题格式YYYY-MM-DD 语音
  • 文档类型smartcanvas(MDX 格式)

⚠️ 写入方式(必须按此操作,禁止其他方式)

步骤 1:用 write 工具将完整 MDX 内容写入临时文件(如 /tmp/speech_paper_YYYYMMDD.md

步骤 2:用 write 工具创建 Python 脚本(如 /tmp/create_tdoc_YYYYMMDD.py),脚本内容固定为:

import subprocess, json

# 从文件读取内容(禁止用 f-string 拼接内容!)
with open("/tmp/speech_paper_YYYYMMDD.md", "r") as f:
    content = f.read()

args = json.dumps({
    "mdx": content,          # 参数名必须是 mdx,不是 content
    "parent_id": "YUsookchBhki",  # 文件夹 ID,必须传
    "title": "YYYY-MM-DD 语音"    # 文档标题
})

result = subprocess.run(
    ["mcporter", "call", "tencent-docs", "create_smartcanvas_by_mdx", "--args", args],
    capture_output=True, text=True
)
print(result.stdout)
print(result.stderr)

步骤 3:用 exec 工具执行 python3 /tmp/create_tdoc_YYYYMMDD.py

步骤 4:检查返回的 file_idurl,确认写入成功后告知用户。

⚠️ 关键注意事项(血泪教训)

  1. 内容必须先写文件,再读文件传参——禁止在脚本里 f-string 或字符串拼接 MDX 内容,否则特殊字符(引号、反引号、换行)会导致 JSON 解析失败或内容截断
  2. 参数名是 mdx,不是 content——传错参数名腾讯文档返回 400001(content is empty)
  3. parent_id 必须传——不传则文档创建在根目录而非目标文件夹
  4. 禁止 shell 直接拼接大段中文内容——必须通过 Python json.dumps 序列化

文档结构模板

# YYYY-MM-DD 语音论文速递

**共收录**: XX 篇 | **语音大模型**: XX 篇 | **语音前端**: XX 篇

> 今日 arXiv 语音相关论文(eess.AS / cs.SD / cs.CL)共命中 XX 篇。

---

## 🤖 语音大模型

[各篇论文内容]

---

## 🎙️ 语音前端

[各篇论文内容]

---

*由开心果 🍀 自动生成 · 数据来源:arXiv*

注意事项

  • 过滤严格执行,宁少勿滥,不混入非语音论文
  • 全程中文输出,论文标题保留英文原文
  • 腾讯文档写入失败时直接在对话输出结果并告知原因
  • 若文档内容过长(> 8000 字),拆为两个文档分别写入,标题加 (上) / (下) 后缀

Comments

Loading comments...