personalized_Podcast_Mandarin

使用此 skill 当用户想要将 URL、PDF 或任意主题生成带有双声道 TTS 音频的 双人对话播客。适用于双人讨论节目、指定主持人 persona（如鲁豫、罗永浩、 李诞风格）、或将文章与资料转化为可收听音频内容的场景。

This package largely implements what it claims (dual‑host script generation + TTS) but has two practical red flags you should address before use: (1) the skill's README/SKILL.md require API keys (DOUBAO_API_KEY and VOLCANO_TTS_*), yet the registry metadata does not declare them — confirm where you should provision those keys and that you trust the provider; (2) the PersonaExtractor is designed to store long, detailed 'memory_seed' items on disk (200–500 word stories), which can contain sensitive personal information. Recommendations: (a) only run in an isolated/test environment (container or VM) until you audit the code; (b) search the repo for any hardcoded endpoints or secrets, and review src/volcano_client_requests.py and src/volcano_client_* for destination hosts; (c) if you must enable TTS, use scoped service credentials and avoid putting real PII into persona memory (or disable persistence); (d) prefer running without enabling external Sub‑Agent websearch until you trust the skill's origin; (e) contact the publisher/owner (none/unknown here) or obtain a vetted release before using with real data.

Type: OpenClaw Skill Name: personalized-podcast-mandarin Version: 1.0.0 The skill is a comprehensive and well-documented tool for generating dual-host podcasts using ByteDance's Doubao (Ark) LLM and Volcano TTS services. It includes legitimate modules for web scraping (web_scraper.py), PDF text extraction (pdf_parser.py), and persona management (persona_manager.py). API keys and configuration are handled through standard environment variables or local files in a 'private/' directory, with a setup wizard provided for user convenience. All network communication is directed to official ByteDance endpoints (volces.com and bytedance.com) or user-specified URLs for content retrieval. No evidence of data exfiltration, malicious execution, or prompt injection attacks was found.