跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
视频与音频

media-scripter

Recorded-media scripting for university professors. 4-agent team turning lecture notes and flipped-class specs into mini-lecture video scripts, shot-by-shot storyboards, 6–9-minute episode series, cleaned captions/transcripts, and podcast-style audio adaptations. Scripts are spoken language with built-in accessibility — not read-aloud prose. Triggers on: video script, record a lecture, lecture video, mini-lecture, screencast, storyboard, captions, transcript, subtitles, podcast, flipped video, MOOC, 录课, 慕课, 视频脚本, 微课, 录屏, 字幕, 讲稿, 播客, 翻转课堂视频.

YujxZJCN · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add YujxZJCN/teaching-skills --skill media-scripter

技能说明


name: media-scripter description: "Recorded-media scripting for university professors. 4-agent team turning lecture notes and flipped-class specs into mini-lecture video scripts, shot-by-shot storyboards, 6–9-minute episode series, cleaned captions/transcripts, and podcast-style audio adaptations. Scripts are spoken language with built-in accessibility — not read-aloud prose. Triggers on: video script, record a lecture, lecture video, mini-lecture, screencast, storyboard, captions, transcript, subtitles, podcast, flipped video, MOOC, 录课, 慕课, 视频脚本, 微课, 录屏, 字幕, 讲稿, 播客, 翻转课堂视频." metadata: version: "1.0.0" last_updated: "2026-06-10" status: active pipeline_stage: 2 related_skills: - lesson-builder - deck-studio - teaching-pipeline

Media Scripter — Recorded Teaching Media Team

Scripts what the professor says into a camera or microphone: lesson-builder decides what the pre-class material covers; this skill scripts the actual videos and audio. Recorded media obeys different rules than a live lecture — there is no room to read, the student has a pause button, and attention decays fast — so scripts are tighter, segmented, and signposted in ways live notes never need to be. The professor brings the content and their own voice; this skill brings spoken-register craft, the recorded- media evidence base, and accessibility built in from the first draft.

Prime rule: lecture notes are raw material, not the script. Prose that reads well silently dies on camera. Every script passes the read-aloud test before it reaches the professor — and every script ships with its transcript, because captions are planned at the start, not bolted on after recording.

Quick Start

Write a video script for my Week 5 flipped class on hash tables
帮我把这章内容拆成几个微课视频,每个不超过 8 分钟
Storyboard a screencast demo of debugging a segfault in gdb
Clean up this auto-generated transcript — the terminology is mangled
Turn my recorded lecture into a podcast episode students can listen to commuting

Modes

ModeTrigger intentOutput
script"Write a video script for…" — one mini-lectureTwo-column script: narration + visual cues + timing, from templates/script_template.md
storyboard"Storyboard / plan the shots for…" — screencasts, demos, visual-heavy segmentsShot-by-shot plan: what's on screen, what's said, screen-action cues, production notes
series"Break this topic/week into videos" — more than one episode's worthEpisode sequence (6–9 min each) with per-episode objectives and the retrieval question between episodes, from templates/series_map_template.md
captions"Fix these captions / clean this transcript" — raw auto-transcript in handAccurate caption file + readable transcript document: terminology fixed against course materials, sentence-segmented, speaker/visual annotations
audio"Make a podcast version / audio-only"Podcast-style adaptation: visual content translated to narration or explicitly deferred, with chapter markers

Mode dispatch rule: "record my Week N video" with no script source → check passport artifacts for a lesson-builder W<N>_preclass_spec.md or W<N>_lecture_notes.md first; none found → offer lesson-builder for content design, or run script with direct intake if the professor already has notes. Material that won't fit one episode → series proposes the split before any script is drafted. Detect intent in any language.

Does NOT trigger

ScenarioUse instead
Deciding WHAT the pre-class material should cover — the flipped design itselflesson-builder (flipped mode)
The slides shown in the video — theme, rendering, figuresdeck-studio
Lecture notes for a live class meetinglesson-builder (lecture-notes mode)
Full design → materials → assessment run across stagesteaching-pipeline

(The boundary in one line: lesson-builder specifies the material and writes the prose; deck-studio renders what's on screen; this skill scripts what's said over it and when. A flipped spec or lecture notes arriving from lesson-builder are consumed as confirmed content — script_writer transforms the register, never the claims.)

Agent Team (4)

AgentRole
script_writer_agentTransforms prose source into spoken-register two-column scripts: signposting, worked-example pacing, embedded retrieval prompts, timing from word count, [VERIFY] discipline on domain claims
storyboard_agentShot-by-shot plans: screen/narration/duration tables, screencast cursor-and-zoom discipline, demo error-recovery planning, talking-head alternation points, production-effort honesty
segmenter_agentTopic → episode architecture: dependency ordering, one objective per episode, sizing from word counts, inter-episode retrieval questions, series map fed to the passport schedule
transcript_editor_agentRaw auto-transcript → caption file + readable transcript: terminology corrected against course materials, caption line conventions, non-speech annotations, marked gaps over confident guesses

Workflow (script mode)

Phase 0  LOAD        — locate the source: lecture notes, slide outline, or flipped
                       spec from passport artifact_refs; else direct intake (the
                       professor's notes/outline + target audience + register sample
                       if available). No source at all → this is content design:
                       route to lesson-builder before scripting.
Phase 1  SCOPE       — segmenter sizes the source against the 6–9-minute default
                       (references/video_pedagogy.md): more than ~9 minutes of spoken
                       material → propose a `series` split instead of one long script,
                       with the episode map. One episode's worth → proceed.
         🧑 checkpoint: scope confirmed (si
阅读完整来源 ↗