ai-research-podcast

当用户说：- "把这篇研报转成音频"- "帮我听听这篇文章"- "生成播客版本

Security Scan

The skill's requests and runtime instructions are consistent with its stated purpose (convert articles/papers to podcast audio); no unrelated credentials or surprising installs are declared, but you should inspect the referenced skills and note a few minor inconsistencies (model download vs. 'offline' claim, push-channel credentials, and optional package installs mentioned in README).

Purpose & Capability

Name/description (convert research articles to audio) align with the instructions: fetch/summarize content, synthesize TTS, optionally push the resulting file. The skill delegates work to three plausible components: summarize, sherpa-onnx-tts, and message.

Instruction Scope

Instructions stay within the stated purpose: they fetch content from URLs or local files, produce a markdown summary, run local TTS to produce audio, and optionally push via Feishu/WeChat/email. Two notes: (1) README and SKILL.md claim '离线' TTS but also state sherpa-onnx requires an initial ~100MB model download — that means a network operation on first run (minor inconsistency). (2) The skill mentions scheduling/cron examples (daily automatic push) but does not include explicit commands; scheduling implies persistent tasks which require agent/system permissions.

Install Mechanism

This is instruction-only (no install spec) so the skill itself won't write code to disk. However README suggests users install summarize via npm and sherpa-onnx via pip/brew; those external installs are expected for the described behavior but are not enforced/managed by the skill. No download-from-untrusted-URL installs are present in the skill bundle.

Credentials

The skill declares no required environment variables or credentials. It references push channels (feishu, wechat, email) but does not declare their credentials — presumably the built-in message skill or platform will handle auth. Before use, verify the message skill or platform will prompt for/securely store any required tokens (Feishu/WeChat/API keys).

Persistence & Privilege

always is false (normal). The workflow includes a scheduled trigger (cron) and examples of setting daily automated tasks—these imply persistence (scheduled runs) but do not by themselves change agent privileges. Confirm that creating scheduled tasks (cron) is acceptable in your environment and that the agent will request permission to schedule jobs.

Assessment

This skill appears coherent for turning articles/URLs/PDFs into short podcast audio: it fetches content (via the summarize skill), converts the summary to speech with an offline TTS (sherpa-onnx), and can push results via the built-in message skill. Before installing/using it: 1) Inspect the referenced skills (summarize, sherpa-onnx-tts, message) to see what network access, model downloads, or credentials they require. 2) Expect an initial model download (~100MB) despite the 'offline' claim; if you need strictly zero-network operation, verify the model is preinstalled. 3) Confirm how push channels are authenticated (Feishu/WeChat/email) and where those credentials will be stored. 4) If you plan to enable the scheduled trigger, confirm the agent is allowed to create cron/scheduled jobs on your system. If you want extra caution, test with local files and disabled push before enabling automation or scheduling.

AI 研报速读播客

将研报、长文、技术文档自动转化为可收听的播客音频。

触发条件

当用户说：

• "把这篇研报转成音频"
• "帮我听听这篇文章"
• "生成播客版本"
• "URL 转语音"
• "研报速读"

执行流程

1. 内容抓取与摘要

使用 summarize 技能：

summarize "<source>" --max-length 500 --output /tmp/summary.md

支持的输入源：

• HTTP/HTTPS URL
• 本地 PDF 文件
• 本地 Markdown/文本文件
• 直接粘贴的文本

2. 语音合成

使用 sherpa-onnx-tts 技能：

sherpa-onnx-tts \
  --input /tmp/summary.md \
  --output /tmp/podcast.mp3 \
  --voice zh_CN-female \
  --speed 1.0

可选参数：

• --voice: 音色选择
• --speed: 语速 (0.5-2.0)
• --output-format: mp3/wav

3. 推送通知（可选）

如果用户指定推送渠道：

# Feishu
message --target feishu --file /tmp/podcast.mp3 --text "AI 研报速读"

# 微信
message --target wechat --file /tmp/podcast.mp3

示例对话

用户: 帮我把这篇论文转成音频 https://arxiv.org/abs/2401.12345

Agent:

1. 抓取论文内容并生成摘要...
2. 将摘要转为语音播客...
3. 完成！音频文件：/tmp/podcast_2401.12345.mp3 (时长 3:42)

用户: 每天早上 8 点自动推送最新的 AI 论文音频

Agent: 好的，已设置定时任务。每天 8:00 会自动：

1. 抓取 arXiv cs.AI 最新论文
2. 生成摘要并转音频
3. 推送到你的飞书

注意事项

• sherpa-onnx 需要首次下载语音模型（约 100MB）
• 离线模式，无需联网，隐私安全
• 建议摘要长度控制在 300-800 字，音频时长 2-5 分钟