Many people assume making TikTok shopping videos means hiring creators or buying a stack of tools. In reality, if you already use OpenAI Codex, you just install one skill and it runs the whole flow for you: find winners → decode them → replicate them → ship the video.
This tutorial covers three things in detail: how to use Codex, how to install the Clipcat skill in Codex, and how to use Clipcat's 5 core scenarios to make publish-ready TikTok shopping videos. Every scenario comes with a copy-paste prompt template.
The short answer
- Codex is a workflow-executing terminal agent; skills live in
~/.agents/skills/. - Install Clipcat in three steps: send the install script → get an API key → configure with
clipcat config. - 5 selling scenarios: viral replication, product-image-to-video, AI find-then-replicate, fixed character, daily automation.
- All driven by plain-language prompts — no coding; free users can trial-generate on the base models.
1. First, understand how Codex works
Codex isn't a chat box — it's a terminal agent that actually acts: it runs commands, loads skills and calls external services. Three concepts get you started.
- Skills directory: Codex stores skills in
~/.agents/skills/— the source of what it can do. - List skills: type
/skillsin a session to see what's available. - Pin a skill: @mention a skill to pin it to the current task so Codex uses it first.
- Standing guidance: write rules Codex should always follow in
AGENTS.md(orAGENTS.override.mdunder~/.codex).
Because Agent Skills is an open standard, the same skill definition also works in Claude Code (~/.claude/skills/), Gemini CLI and Cursor. Learn it once, use it everywhere.
2. Install and configure Clipcat in Codex
Install the skill
Send the line below plus the install script to Codex and it installs automatically. macOS/Linux and Windows use different scripts — pick the right one.
# macOS / Linux
Please install the Clipcat skill.
curl -fsSL https://clipcat.ai/cli | bash
# Windows
Please install the Clipcat skill.
irm https://clipcat.ai/cli.ps1 | iexGet an API key
Sign up or log in at clipcat.ai and generate an API key in the personal center. Keep it safe — it's used for API and agent service calls.
Configure the API key
Replace your_api_key with your own key in the command below and send it to Codex. Once configured, Clipcat is callable from any agent.
clipcat config --api-key your_api_key --base-url https://clipcat.aiClipcat is published on Clawhub and GitHub, where you can also download it.
3. Create shopping videos with 5 scenarios
These 5 scenarios cover the highest-frequency ways to produce TikTok shopping content. Each has a copy-paste prompt template.
About the {} in each template: anything wrapped in curly braces (e.g. {URL}, {product image}, {the US}, {15s}, {seedance2.0}) is a replaceable variable — swap the braces and their sample together for your own link, product, target market, duration and model. Don't keep the braces themselves.
Scenario 1: Viral replication — one TikTok link → a new video
See a great video? Drop the TikTok link to Codex. It captures the winner's script structure and shot logic, and generates a new, publish-ready version for your product on the same proven framework.
Prompt template
Use the Clipcat skill's replication feature to fully replicate the original script and shots, only swapping the product for mine (adjust voiceover to my product; none needed if there isn't any), no subtitles, keep the new script within {15 seconds}, model {seedance2.0}, voiceover in {English}. Here is the TikTok link to replicate: {URL}
- URL: must contain a video id — copy from the PC web version (usually includes the id).
- Product images: required, up to 9, total under 10MB.
- Spec: default 720P/15fps/15s, adjustable in the prompt. Free users can trial on the base models; just ask your agent for the exact credit cost.
Scenario 2: Product-image → video — build local viral shopping clips
Input product images and a requirement to auto-generate a native-voiceover shopping clip. Multi-angle images improve product consistency (the system auto-builds a white-background 9-grid for the model), and one person can recur on screen.
Prompt template
Using the Clipcat skill, take this product {product image} and generate a {prompt, e.g. UK girl OOTD outfit video}, duration {15s}, model {seedance2.0}, voiceover in {English}.
- The requirement matters most: clearly state the desired result (scene, character, selling point).
- Defaults: 15s, English voiceover — adjust in the prompt.
Scenario 3: AI finds TikTok winners, then replicates them
One of the most popular scenarios: it chains search → shortlist → analyze → replicate into one pipeline, saving 3–5 hours/day of manual watching and script breakdown while producing fresh, on-trend assets.
Prompt template
Use the Clipcat skill to find the hottest shopping videos for {product category} in the last 7 days in {the US}, then — given my product image — pick the best-matching winner and get its TikTok link. Then use the replication feature to fully replicate the script and shots, only swapping in my product, no subtitles, keep within {15s}, model {seedance2.0}, voiceover in {English}.
Scenario 4: Videos with a fixed on-screen character
Generate or upload a photo to create a digital persona that recurs across all your videos — no shooting, no real person. It keeps one consistent look and style so your account builds a recognizable identity.
Prompt template
My on-screen character is {character id}, and {product image} is my product. Use the Clipcat skill to fully replicate the original script and shots, only swapping in my product (adjust any voiceover; none needed if absent), no subtitles; if the original has a person, the new one should too. Keep within {15s}, model {seedance2.0 mini}, voiceover in {English}. Here is the TikTok link: {tk link}
- Character id: get it from Clipcat's "My Characters" page, or provide an accessible character image URL.
Scenario 5: Daily auto-find-and-generate — a content perpetual machine
Set a scheduled task once, and Codex finds a fresh winner in your niche every day and replicates it onto your product — new videos waiting for you each morning. Keep chatting to iterate the next batch.
Prompt template
Set a scheduled task: {every day at 9am} use the Clipcat skill to find the hottest {product category keyword} shopping videos in the last 7 days in {the US} (not repeating past references). Given my product image, pick the best match and get its TikTok link, then fully replicate the script and shots, swap in my product, no subtitles, keep within {15s}, model {seedance2.0 mini}, voiceover in {English}. Don't ask me to confirm each run — generate directly, log results into one sheet, and send me the new video.
With a scheduled task, your role shifts from "making assets" to "reviewing assets": every morning you have new shopping videos replicated from the latest winning logic.
FAQ
What is Codex and how is it different from a chatbot?
Codex is OpenAI's terminal agent that actually executes workflows — running commands, loading skills, calling external capabilities. It stores skills in ~/.agents/skills/, lists them with /skills, pins one to a task with @mention, and takes standing guidance from AGENTS.md.
How many steps to install Clipcat in Codex?
Three: (1) send "Please install the Clipcat skill" plus the install script to Codex; (2) get an API key in the clipcat.ai personal center; (3) run clipcat config --api-key your_api_key --base-url https://clipcat.ai. Then any agent can call it.
What are the requirements for the TikTok link when replicating?
The link must contain a video id — copy it from the PC web version (those usually include the id). You must upload your product image(s): up to 9, total under 10MB. Default output is 15s with English voiceover; adjust in the prompt.
Which video models are supported?
Grok1.5, Veo3.1, omni, Seedance2.0, MinimaxH3, Seedance2.5 and more. Free users can run trial generations on the base models to get going. The exact credit cost varies dynamically by model, length and resolution — just ask your agent and it returns the current cost.
Where do I get the character id for a fixed character?
Get the id of an existing character from Clipcat's "My Characters" page, or provide a publicly accessible character image URL. Once set, the same digital persona recurs across your series.
Do failed generations still consume credits?
If a task fails before it truly starts processing, credits are refunded automatically; tasks that fail after queuing are also returned to the corresponding API key's balance.