Pluviobyte/rnskill

ra-hands-on-planning

Goal

AI acts as the planning copilot, while the user is the on-screen lead. Produce a planning script that is 'ready to record': test sets, structure, voiceover script, and operation checklist are all prepared, with deliberate blank space left for on-the-spot reactions (the 'human touch preservation zone').

Full Production Line (Planning Script Status State Machine)

Planning (Claude) → Recording (You) → Rough Cut (Codex · Tool Selection Gate) → Merging (You · CapCut) → Subtitles (Codex · ra-audio-to-subtitles) → Animation (Codex · Remotion) → Final Cut → Published

• Raw Footage: Two streams: voiceover + screencast → 01-ContentProduction/VideoWorkbench/InProduction/YYYY-MM-DD-Topic/RawFootage/
• Rough Cut Delivery: <Project>/RoughCut/Voiceover-RoughCut.mp4 (Pure MP4, feed to CapCut)
• Merging (Manual by User): Full-body segments on full screen; all other voiceover segments shrunk and placed in the lower-left PiP of the screencast footage.
• Subtitles: The merged video serves as the sole time source, going through ra-audio-to-subtitles to generate real word-level timestamps, burn hard subtitles, and keep the .srt file (for YouTube CC / Bilibili subtitle script).
• Animation: Only full-body segments get Remotion component overlays; PiP segments do not.
• Distribution: Main publish on Bilibili + YouTube, full distribution on Douyin, Xiaohongshu, WeChat Channels; final cut → VideoWorkbench/Completed/<EarlyMonth|LateMonth>/YYYY-MM-DD-Topic/FinalCut/ + calendar registration.

Planning Script Structure Methodology (Hands-On Comparison Video Paradigm)

1. Hook (within 10 seconds): Hot event + conflict question + promise ("X just released, is it really better than Y from a few days ago? Today we'll compare from three angles: A/B/C").
2. Preemptive Hot Take: Drop a counter-intuitive, conclusive feeling first as a suspense hook ("For the same task, it was 10x faster") to keep the audience watching.
3. Test Suite × N: Fixed loop per question: Read the question (prompt text on screen) → Watch and comment on Result A → Result B → Single-question verdict. Questions must have a gradient design: from "Does it meet the requirements?" to "Competing on a single dimension". The dimensions should have broad coverage (e.g., one question for frontend, one for backend, one for copywriting).
4. Mid-session Summary + Human Standpoint Sentence (a genuine statement like "The results are what they are, I can't lie about it").
5. Conclusion: Verdict by dimension, no fence-sitting.
6. Next Episode Hook: Leave one untested dimension ("Agent capabilities need long-term testing; I'll keep reporting").
7. Ending Easter Egg: Interactive suspense ("Guess who made the thing at the start of the video?").

Hard Requirements for the Planning Script

• Generate using the assets/planning-template.md from this Skill, one project per video: 01-ContentProduction/VideoWorkbench/InProduction/YYYY-MM-DD-Topic/planning-script.md
• Write the voiceover script in two layers: separate the full-body segments (opening, mid-session summary, conclusion – segments that will get Remotion components) from the PiP segments (key points for the voiceover during the test). Write the full-body segment list in the frontmatter for animation team positioning.
• Provide a copy-paste ready full prompt for each test question + one sentence on "What to look for in this question" (expected highlights / potential failure points).
• Don't script the live reactions: The voiceover only provides key points. Leave the commentary and emotional reactions for you to perform live – this is the source of human credibility. Every sentence must not be fully written out.
• The screencast operation checklist must be precise down to the step (what tool to open, what to type, which screen to record).
• Video length controlled by the frontmatter contract; 8-15 minutes default, 4-6 questions.
• For reference videos, only extract the structure and question types; copying original sentences or specific questions from reference videos is prohibited.
• After planning is complete, create a backlink to the topic card (create one if it doesn't exist), and change the card status to 'Planned'.

Handoff with the Production Team

• You only change the status and place raw footage; everything else is automatic: say so when recording is complete or change status to status=Rough Cut Ready; when the CapCut merge is complete and the file is back in the project folder, change status to status=Subtitles Ready.
• Codex's responsibilities are listed in ra-video-production-director items 11-12; when live action footage is ready, ra-local-talking-head-cut is used by default: First deliver the pre-edit proofread SRT to the user for confirmation. Block editing until confirmation. After confirmation, deliver a rough cut MP4 that has passed natural rhythm, source parameter, loudness, and QC checks, and regenerate the production subtitles for the final MP4. For multiple takes, use video-use to provide evidence for the best selection; use chengfeng for word-by-word review or aggressive comparison mode; when waveform review is needed and the project needs to be handed to CapCut/FCP, use AI-cut-voiceover to export FCPXML. Blocking of any one rough cut tool does not affect subtitles proceeding via their own independent Skill.