English | 简体中文
An Agent Skill for turning talking-head, explainer, and screen-recording videos into reviewed final MP4s. The workflow is deliberately staged: clean the spoken base video first, discuss the HyperFrames motion plan with the user, create example frames/pages for confirmation, then render and verify the final video.
These frames come from a real workflow demo made with this skill's process: clean the spoken base, discuss the overlay direction, confirm example frames, then render the final MP4.
- Cleans the spoken base cut: filler words, repeated phrases, false starts, half-spoken English tool names, and no-speech gaps over the requested threshold.
- Reports the cleanup result before motion work: preview path, duration change, removed categories, and uncertain review points.
- Plans the motion layer before implementation: visual direction, overlay coverage, subtitle behavior, and anti-patterns.
- Requires example pages or frames before full production, so the user can approve the style before the full HyperFrames build.
- Renders a final MP4 and verifies it with actual output-file checks, including
ffprobemetadata and representativeffmpegframe extraction.
.
├── skill/
│ └── video-use-hyperframes-editor/
│ ├── SKILL.md
│ ├── agents/openai.yaml
│ └── references/workflow-checklists.md
├── dist/
│ └── video-use-hyperframes-editor.skill
├── docs/
│ └── showcase/
├── scripts/
│ ├── validate_skill.py
│ └── package_skill.py
├── examples/
│ └── prompts.md
└── .github/workflows/validate.yml
Copy the skill folder into your Codex skills directory:
Copy-Item -Recurse .\skill\video-use-hyperframes-editor "$env:USERPROFILE\.codex\skills\video-use-hyperframes-editor"Or install from the packaged .skill file if your client supports skill imports:
dist/video-use-hyperframes-editor.skill
You can also keep the skill inside a project-level skills/ directory:
your-project/
└── skills/
└── video-use-hyperframes-editor/
Use $video-use-hyperframes-editor to process this talking-head screen-recording video.
First clean filler words and repeated phrases carefully, report the cleanup result,
then propose HyperFrames motion directions and create example pages for confirmation
before rendering the final MP4.
More examples are in examples/prompts.md.
python .\scripts\validate_skill.py .\skill\video-use-hyperframes-editorpython .\scripts\package_skill.py .\skill\video-use-hyperframes-editor .\distThe output is:
dist/video-use-hyperframes-editor.skill
- Base cut first, motion later.
- ASR cleanup must be clause-level, not just keyword search.
- User confirmation gates are part of the workflow, not optional politeness.
- Motion overlays should explain structure, not repeat subtitles.
- Chinese overlay text should not use per-glyph stroke outlines.
- Final MP4 acceptance depends on rendered-file checks.
MIT. See LICENSE.


