cs-auto-videl
Use when the user types $cs-auto-videl or asks for Douyin/TikTok ecommerce video workflows, including competitor replication, strong-Hook nine-grid prompt_image.md to image2 storyboards, Seedance prompts/API submission, Gemini Omni first-frame video, or Google Flow/Veo first-frame-to-video packages.
ChenShuo2004 · AI review on file
Installation
Read the source and dependencies before running a command. Copying does not install or execute anything.
npx skills add ChenShuo2004/cs-skills --skill cs-auto-videlAbout this skill
name: cs-auto-videl description: "Use when the user types $cs-auto-videl or asks for Douyin/TikTok ecommerce video workflows, including competitor replication, strong-Hook nine-grid prompt_image.md to image2 storyboards, Seedance prompts/API submission, Gemini Omni first-frame video, or Google Flow/Veo first-frame-to-video packages." metadata: author: "陈硕" collection: "CS Skills" source: "https://github.com/ChenShuo2004/cs-skills" compatibility: "Codex and any agent that supports SKILL.md"
auto Videl
This is the user's local CS Auto Videl skill, invoked as $cs-auto-videl.
It runs Douyin/TikTok ecommerce information-feed video workflows. Use it when the user wants to analyze a benchmark short video, replace the product, generate actual storyboard images from prompt_image.md with image2, turn user-provided nine-frame storyboards into a final video, generate Seedance 2.0 prompts, prepare Gemini Omni or Google Flow/Veo first-frame-to-video packages, optionally submit Seedance API generation tasks with their own locally configured API key, or consolidate results into Feishu.
Platform Mode Selection
Before writing final video-generation deliverables, identify the target platform mode. If the user has not named a mode, ask which mode to use before producing final prompts or images:
seedance: Seedance/C端2.0 prompts, storyboard references, optional API dry-run or submission.gemini-omni: 8-second first-frame-to-video units with first-frame images and Gemini Omni prompts.google-flow: Google Flow/Veo first-frame-to-video units, normally 8 seconds each, with first-frame images, image prompts, video prompts, and optional voice/sound direction.prompt-only: no image/video generation; deliver scripts, storyboard prompts, and platform-ready text only.
If the user explicitly says $cs-auto-videl google-flow, Google Flow, Flow, Veo, Frames to Video, or 首帧转视频, load references/google-flow-mode.md and use Google Flow mode. For Google Flow mode, default to first-frame-to-video unless the user explicitly asks for another Flow feature. If the user asks for Google Flow but not the clip duration, use 8 seconds per clip unless they ask for a supported shorter duration.
Route Selection
Before starting, choose exactly one route from the user's input.
- 复刻链路: use the benchmark video as the source of shot order, timing, scenes, actions, copy rhythm, and composition. Generate storyboard images first, then generate Seedance prompts, and only call the API when the user's private key is configured.
- 强 Hook 九宫格生成出片链路: use when the user gives product information, selling points, target audience, product images, or says to use
prompt_image.md/Codex instead of Gemini to generate nine-grid prompts. Codex reads bundledreferences/prompts/00-prompt-image.mdor the workspaceprompt_image.mdwhen the user explicitly points to it, generates a strong-Hook nine-grid Chinese storyboard script, waits for user confirmation, generates 9 image2 prompts, calls image2 to create 9 storyboard frames, then writes the Seedance/C端2.0 prompt and optionally calls Seedance. The expected handoff isprompt_image.md→ image2 → Seedance, with 9帧分镜图 + 9段提示词 retained as core artifacts. - 九宫格成片直投链路: use when the user uploads 9 storyboard frames and 9 prompts that were already produced from the
prompt_image.mdsystem. Skip script generation and image2. Use the provided 9帧分镜图 + 9段提示词 to write the final Seedance/C端2.0 prompt, then optionally submit it through the API. - If the input contains a benchmark/competitor video, route to 复刻链路. If the input lacks a benchmark video and asks Codex to create the nine-grid prompts/images, route to 强 Hook 九宫格生成出片链路. If the input already contains 9 frames plus 9 prompts, route to 九宫格成片直投链路.
- If multiple route signals are present, ask which route to use unless the user explicitly names one route.
- Never mix routes inside one run. A replication run may produce storyboards, a generation run creates storyboards from
prompt_image.md, and a direct-submit run starts after the 9 frames and 9 prompts already exist.
Core Contract
- Product fidelity is the P0 gate. If the replacement product is wrong, deformed, missing required identity details, or not faithful to the product images, the run fails even if the composition looks good.
- Product images lock product appearance, structure, color, material, texture, logo/mark position, logo/mark size, logo/mark color, logo/mark direction, and any visible product-specific details.
- Do not solve logo/text risk by removing or weakening the product identity. If the product has a visible logo/mark in the product images and the benchmark shot angle should show that product area, the storyboard must preserve it in the physically correct position and scale.
- Storyboard images are mandatory final deliverables for replication runs unless the user explicitly asks for prompts only. For 强 Hook 九宫格生成出片链路, image2-generated 9-frame storyboards are mandatory. For 九宫格成片直投链路, the user-provided 9 frames are the storyboard images and do not need to be regenerated unless the user asks.
- Use Codex's built-in GPT image/image2 generation capability to create the storyboard images; do not stop at storyboard prompt text.
- Storyboard images lock visual structure, shot order, scene continuity, product placement, and repeated people/hand/background style.
- Original video breakdown locks timecodes, shot sequence, actions, and scene details.
- Rewritten copy locks voiceover rhythm and conversion path.
- Final video prompts must use concrete director language.
- Storyboard image prompts and Seedance video prompts must obey real physical-world logic: plausible gravity, scale, hand/object contact, body movement, reflections, lighting, camera position, and continuity.
- For videos clearly longer than 17 seconds, consistency across multiple s