Communitygithub.com

mmsx4717/news-video-generator

Generate or modify news videos using RSS, MiniMax, and Remotion, with fact-checking, enterprise grouping, screenshots/demo, voiceover, subtitles, and local export—suitable for AI and other news domains.

What is news-video-generator?

The `mmsx4717/news-video-generator` skill enables AI agents—particularly those like Codex with file, command, and browsing capabilities—to autonomously produce or edit news videos from RSS feeds using MiniMax and Remotion. It supports a complete workflow: topic selection, script drafting, fact verification (via real screenshots and evidence checks), enterprise grouping for coherent narrative flow, voiceover with adjustable speech rate (1.1), subtitle generation, and final local video export. The skill emphasizes journalistic integrity: all claims must have factual basis, news status and limitations must not be altered, and exaggerated expressions require evidence. It integrates with the user's existing project structure (docs/workflow.md, docs/data.md, docs/production.md) and avoids automatic installation on other agents. Key safeguards include not copying personal data (voice samples, credentials) to version control, verifying credentials before calling paid TTS (dry runs are free), and requiring actual inspection of screenshots rather than treating success as confirmation. The skill is designed for targeted, user-directed news production rather than bulk automation or scheduled tasks.

Works with~Claude Code✓Codex CLI~Cursor
npx skills add https://github.com/mmsx4717/news-video-generator/tree/HEAD/skills/news-video-generator

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

News Video Generator

Run this skill within the repository using Node.js 22+. It does not depend on personal directories or other skills. Suitable for Codex and agents with file, command, and browsing capabilities. Input: user's topic, RSS feed list, and production scope. Do not automatically install on other agents.

  • Topic Selection, Script Drafting, Fact-Checking: Read editorial workflow. Perform a single collection from the user-specified RSS feeds. Do not introduce forced aggregation platforms or scheduled two-round processes.
  • Input and Caching: Read data format. Reuse stable IDs and unchanged valid materials. Preserve original evidence and new details.
  • Collection, Screenshots, Voiceover, Assembly, Preview, and Delivery: Read production commands and the README command examples in the repository.

"Selection" implies reducing the number of news items. For included products/models, clearly explain their actual use, changes, examples, and availability conditions. Narrate naturally without reading a fixed outline. Arrange content by enterprise continuity. Exaggerated expressions must have a factual basis. Do not change the status or limitations of the news. Set speech rate uniformly to 1.1, and insert pauses between sentences based on semantics.

First verify existing dependencies and the current state of runs/. Report any missing credentials without printing keys. Only call paid TTS when explicitly producing and configured with a voice; dry runs do not incur charges. Do not copy user images, voice samples, subscription credentials, or historical scripts into version control. The default generic template does not include a personal persona.

Screenshot success does not equal completion of fact-checking. Inspect the actual images and record receipts. During review, examine every sentence, transitions, and keyframes at the start and end. Claim full audio proof only when it has actually been performed. If the user requests a preview only, do not export. After export, perform only local delivery; do not attach uploads, publishing, or scheduled tasks.

Related Skills