Introduction

In the DeepSeek Harness (DSH) ecosystem, a Skill is usually responsible for interpreting instructions and scheduling tools. For video editing tasks that involve multiple local tools (transcoding, speech recognition, and audio synthesis), manual configuration and scheduling can be very cumbersome. This plugin packages the local open-source toolchain into a Skill, so a single natural-language command can take over the entire process from source material analysis to final delivery.

Plugin Overview

This is a workflow-orchestration Skill maintained by user liuke-zhu. It does not perform editing directly; instead, it understands user requirements and schedules local tools such as FFmpeg, Faster-Whisper, and VoxCPM to complete the specific work.

Core Capabilities

  1. Workflow orchestration: Understands requirements and schedules tools, executing editing tasks step by step.
  2. Local processing: Source material does not leave the computer. Uses Faster-Whisper for transcription and VoxCPM for voiceovers, without calling cloud services.
  3. Natural-language editing: Speak in plain language to modify the video. For example, “make the beginning faster” or “make the subtitles larger”; deleted clips can be recovered from the original material.
  4. Conservative editing: Automatically removes obvious silences, repetitions, and mistakes; clips with uncertain meaning are kept by default, and the original material is never modified.
  5. Delivery self-check: Before delivery, uses FFprobe to verify the output file’s validity, readability, aspect ratio, and duration.

Installation and Enablement

This plugin supports the DSH plugin protocol and can be installed with dsh plugin add, or by manually cloning the repository.

Method 1: Install via the DSH plugin protocol

dsh plugin add

After installation, the plugin registers the “Really Love You” skill with the Agent as a bundled skill.

Method 2: Manual installation

git clone https://github.com/liuke-zhu/zhenxin-ai-video-manager
cd zhenxin-ai-video-manager
.\install.ps1

install.ps1 copies SKILL.md to ~/.dsh/skills/zhenxin-ai-video-manager/.

Prerequisites

This Skill does not include the editing toolchain; it must be installed separately:
* FFmpeg / FFprobe: Cutting, concatenation, transcoding, and subtitle burn-in.
* Python 3.11: Runtime environment for scripts.
* Faster-Whisper: Local speech recognition.
* VoxCPM: Local AI voiceover.
* pyJianYingDraft (optional): Generating JianYing drafts.

Quick Start

  1. Install the toolchain (such as FFmpeg, Python, Faster-Whisper, and VoxCPM).
  2. Install the Skill (run the installation commands above).
  3. Configure the paths. Run .\doctor.ps1 for an environment check; it prints a “SKILL.md path mapping table”. Open SKILL.md, find the “Tool Responsibilities” section, and change paths such as ffmpeg.EXE, faster-whisper-small, and voxcpm-py311 to their actual installation locations.
  4. Restart the session and give the command.

Typical Usage

After installation and configuration, enter the following commands in the Agent session:

  • Start: Really Love You, start creating the video
  • Modification commands: Faster at the beginning / Keep this segment / Larger subtitles / More natural voice / Stop
  • Restore deletion: Restore the part just deleted

Notes

  • Local execution: The editing toolchain (FFmpeg / Faster-Whisper / VoxCPM) must be installed separately. The plugin runs with the permissions of the current DSH process; ensure it has read/write permissions for the tool paths.
  • Material safety: It does not upload source material by default and does not call paid cloud services. Original material is read-only; it is not moved, deleted, or overwritten.
  • Capability boundaries: It does not generate film visuals out of nothing; it only processes real material. It does not automatically publish to any platform. It does not treat “the script has finished” as “quality has been verified”.

Summary

This plugin encapsulates a complex local video editing workflow into a single natural-language command, suitable for users who need quick video production while valuing data privacy. Through a conservative editing strategy and strict delivery self-checking, it ensures the output meets expectations.

  • GitHub: https://github.com/liuke-zhu/zhenxin-ai-video-manager
  • Directory: https://www.skillhub.cn/plugins/liuke-zhu/zhenxin-ai-video-manager