Official Memm skills
Ten free skills that take Claude Code or Codex from understanding a business to finished, checked ad creatives.
Last reviewed:
What the skills do
Ten Memm skills help Claude Code and Codex with the whole job, from understanding a business to finished ad creatives that have been checked against the brief. A skill is a set of working instructions for your assistant, not a new model. The skills work in Arabic and English and use the models available to your account.
| Skill | Use it for |
|---|---|
| memm-start | Start here: checks your Memm connection for free, picks the right skills and keeps the project notes |
| memm-research | Understanding a business, its customers and competitors from public sources |
| memm-business-plan | A focused growth or campaign plan with recommendations you can test |
| memm-workflow-plan | Stages, owners, time, credits and checks |
| memm-creative-strategy | Ad concepts and the reasoning behind each one |
| memm-tool-selection | Which model to use, what it costs and how to spend fewer credits |
| memm-production | Images, product photos, Arabic ads, video, voice-overs, music and 3D |
| memm-lip-sync | A photo that speaks or sings, motion copied from a video, edits to your own video |
| memm-capoeira | Your product, character or outfit inside an existing video, keeping its motion, or a move from the Motion Library on your character |
| memm-qa | Checking every result against the brief before it is final |
Download the skills (ZIP) · Public repository · Version 2.1.0. Installing is free. Generations use your Memm credits, and the assistant shows the price and waits for your yes before it starts.
You do not need to name a skill: ask in your own words and the assistant picks the right one. This version replaces the earlier Product Images and Product Videos skills; their guidance is now part of memm-production.
Install
Windows: unzip the download, open PowerShell in the memm-skills folder and run:
powershell -ExecutionPolicy Bypass -File install\install.ps1
macOS and Linux: from the same folder, run:
bash install/install.sh
The installer finds Claude Code and Codex, installs the skills in each, and adds the Memm connection only if you do not already have one. It backs up your settings before any change and touches nothing else. Run it again to update. Add -Uninstall (Windows) or --uninstall (macOS and Linux) to remove what it added.
Without the installer, in Claude Code:
claude plugin marketplace add activecx/memm-skills
claude plugin install memm@memm
In the Codex app or CLI:
codex plugin marketplace add activecx/memm-skills
codex plugin add memm@memm
The Codex IDE extension does not load plugins: copy the folders from plugins/memm/skills/ into ~/.agents/skills/ instead. If your assistant is not connected to Memm yet, follow Connect your assistant.
Sign in once
In Claude Code, type /mcp, choose memm, then choose Authenticate. In Codex, if no sign-in page opened in your browser during installation, run codex mcp login memm.
Then start a new session and ask: Check my Memm setup. The assistant uses two free tools and reports your plan and balance. No credits are used. Generating needs a paid Memm plan; research and planning work without one.
Try it
Two requests to start with:
Use Memm. Study my business from mysite.com, then suggest three different ad ideas for Instagram, with the reasoning for each. Do not generate anything yet.
Use Memm. Make one square photo of my product from my product image, with no text. Show me the model and the price before you start.
The skills keep short notes for each piece of work in a memm/<project>/ folder inside your working folder: the brief, sources, decisions, budget, jobs and reviews. You do not have to repeat yourself, and every credit spent can be traced. You can delete the folder at any time.
What the skills cannot do
- Post or schedule on social media.
- Run or buy ads, or read their results.
- Dub an existing video in one step.
- Listen to audio, so pronunciation and lip-sync timing always need your own check.
The skills say so instead of pretending. They cannot guarantee results either: every result is checked against your brief, and the assistant tells you what it could not check.
What we tested
Version 2.1.0 was tested in Claude Code and Codex against a simulated Memm server in 13 situations, including price approval, a retry after a timeout, failed jobs, lip-sync permission and Arabic replies. No paid generations were made in testing, so judge output quality with your own approved examples.