Post-production video editing is often where creative momentum dies. You finish a 40-minute recording, and then the real work begins: scrubbing timelines, hunting for quotable lines, and reconstructing the narrative flow. Developer Robin Beraud has built a solution to automate this mechanical burden, releasing a new Hermes skill called transcript-to-chapters on November 19, 2026.
How the Skill Works
The tool operates on a simple premise: it takes the subtitle files (SRT or VTT) that most editors and platforms already export and analyzes them for structural changes. The underlying script has no dependencies and requires no API key, keeping the barrier to entry low. It reads timestamps and compares neighboring stretches of speech to identify where vocabulary shifts significantly, which typically signals a new topic. Additionally, it scans the transcript for sentences that function well as standalone quotes.
The Agentβs Role in Curation
While the script handles the detection, the Hermes agent applies judgment to the output. It generates short, honest chapter titles and adjusts boundaries to align with the sentence that actually starts a new topic. Crucially, it ensures that any extracted quotes are copied word-for-word with their original timestamps, preventing the common issue of misattributed or reworded citations. The tool also validates the output against YouTubeβs specific chapter rules, such as ensuring the first chapter starts at 0:00 and that there are at least three chapters, each lasting a minimum of ten seconds.
Limitations and Current State
Beraud is transparent about the toolβs current capabilities. The chapter boundaries are suggestions based on word count analysis, meaning short or uniform talks may still require manual adjustment. Furthermore, since the tool relies on auto-captions or exported subtitles, it inherits any transcription errors present in those files. The developer notes that the skill has been tested on a small sample and self-tests, but has not yet been validated against hours of real-world recordings. It is available under an MIT license in the hermes-skills library on GitHub.
Key Takeaways
- The skill requires no API keys or external dependencies, making it lightweight and easy to integrate.
- It automatically validates chapter structures against YouTubeβs technical requirements.
- Quotes are preserved verbatim with timestamps to maintain accuracy.
- Current testing is limited to small samples, so manual review is still recommended for high-stakes content.
The Bottom Line
This is a practical, low-friction tool for creators who hate the mechanical side of video editing. It wonβt replace human judgment for nuanced storytelling, but it saves the tedious minutes spent scrubbing timelines.