AgentSkillsCN

mixcut

通过自动化分析、剪辑与专业字幕处理，从 YouTube 视频片段中快速制作混剪短片。

SKILL.md
--- frontmatter
name: mixcut
description: >
  Create short mixed-cut videos from YouTube clips with automated analysis, clipping, and professional subtitle processing.
allowed-tools:
  - Read
  - Write
  - Bash
  - Glob
  - AskUserQuestion
  - Task
model: claude-3-5-sonnet-20240620

MixCut Skill

CRITICAL INSTRUCTION: MANDATORY CONFIGURATION

You MUST NOT proceed to downloading or processing until the Project Configuration is fully defined.

Step 1: The Configuration Gate Check if the user has provided ALL of the following:

  • •Video Title: Main title for intro and segment cards.
  • •Caption Language: Target language for subtitles (e.g., English, etc.). Crucial: No bilingual output.
  • •Aspect Ratio: 16:9 (Desktop), 9:16 (Shorts/TikTok), or 1:1.
  • •Target Duration: 30s, 60s, 120s, or 180s.
  • •Pace: Fast, Medium, or Slow.
  • •Style: Highlight Montage or Beat-cut.
  • •Transitions: none, fade, slide, flash, or glitch.

If ANY of these are missing, you MUST use the AskUserQuestion tool to obtain them. DO NOT assume defaults. DO NOT start downloading until configuration is complete.

Workflow

Phase 1: Sources & Workspace Setup

  1. •Gather Sources: Ask for YouTube URLs from the user.
  2. •Setup Workspace:
    • •Ensure the directory ./work exists in the project root.
    • •Prepare ./work/assets and ./work/outputs subdirectories.

Phase 2: Content Analysis

For each YouTube URL provided:

  1. •Download:
    • •Run python skills/Youtube-clipper-skill/scripts/download_video.py <url> "./work/assets/youtube"
  2. •Analyze:
    • •Run python skills/Youtube-clipper-skill/scripts/analyze_subtitles.py <subtitle_path>
  3. •Selection:
    • •Present analyzed chapters/highlights to the user.
    • •Confirm which segments to include in the final cut.

Phase 3: Assembly & Processing

  1. •Configuration: Create ./work/mixcut_config.json with global settings and source clip metadata (start, end, title).
  2. •Processing:
    • •Run python skills/mixcut/scripts/process_mixcut.py "./work/mixcut_config.json"
    • •This script automates clipping, media normalization (ffmpeg), and generates the initial timeline.json and final.srt.

Phase 4: Professional Subtitle Cleaning (Agent-Driven)

This phase ensures high-quality, professional subtitles.

  1. •Read Timeline: Access ./work/timeline.json.
  2. •Subtitle Optimization: Iterate through all subtitle entries and apply the following cleaning logic:
    • •Remove Noise: Strip speaker markers (e.g., >>, >>>) and sound cues (e.g., [Music], [Applause]).
    • •Deduplication: Remove repeated words, stutters, and verbal fillers ("uh", "um", "you know").
    • •Flow Improvement: Merge fragmented phrases into natural, readable sentences.
    • •Professional Polish: Refine text into the target language using appropriate technical terminology.
  3. •Update: Save the optimized content back to ./work/timeline.json.
  4. •Synchronize: Run python skills/mixcut/scripts/process_mixcut.py "./work/mixcut_config.json" --action generate_srt to sync changes to final.srt.

Phase 5: Rendering

  1. •Delegate to Remotion:
    • •Use the remotion skill to render the project located in skills/mixcut/remotion.
    • •Provide the absolute path to ./work/timeline.json as input props.
    • •Set output to ./work/outputs/final_mixcut.mp4.

Phase 6: Delivery Guard & Verification

Mandatory verification before completion.

  1. •Integrity Check:
    • •Verify ./work/outputs/final_mixcut.mp4 exists and has a file size > 0.
    • •Verify ./work/outputs/final.srt exists and has a file size > 0.
  2. •Troubleshooting: If files are missing or empty, identify the failure point in previous phases and re-run the necessary steps.

Phase 7: Completion & Cleanup

  1. •Report: Provide the paths to the final video and subtitle files.
  2. •Retention Policy: Do NOT delete the ./work directory until the user explicitly confirms the results or you have verified the outputs physically.

Technical Guidelines

  • •Path Management: Always use absolute paths when delegating tasks to external skills.
  • •Cross-Platform: Avoid hardcoding drive letters; use system-agnostic path resolution.
  • •Performance: Use parallel tool calls for independent download/analysis tasks where supported.