BrainrotShorts Agent Skills are reusable instructions that teach compatible AI agents how to create, edit, render, and retrieve vertical short-form videos with BrainrotShorts. Instead of explaining the production process in every chat, you install the skill once and ask for the outcome you want: a dialogue short, a Reddit story video, an AI voice-over, or a finished 9:16 MP4.
The skill is available through its skills.sh listing, and the complete source is visible in the BrainrotShorts Agent Skills GitHub repository.
That small shift—from giving an AI a generic prompt to giving it a production playbook—is what makes agent skills useful. The assistant is no longer improvising the order of API calls or guessing which settings exist. It knows how to discover valid characters and backgrounds, prepare a draft, ask for approval before a paid render, monitor the job, and return the output URL.
What Are BrainrotShorts Agent Skills?
An Agent Skill is a folder containing a SKILL.md playbook and, when needed, supporting references, scripts, or assets. Compatible agents load the short name and description first, then read the full instructions only when a request matches the skill. This progressive approach keeps the assistant's context lean while still giving it specialized knowledge at the right moment.
The BrainrotShorts skill packages three kinds of knowledge:
- Creative workflow guidance: how to shape a hook, script a fast dialogue, structure a Reddit story, or split an AI voice-over into scenes.
- Production procedures: how to create and update projects, select catalog values, apply scripts, place timed overlays, render, poll status, and retrieve an MP4.
- Safety and reliability rules: how to keep credentials secret, avoid invented IDs, separate MCP and REST keys, request approval before renders, and confirm destructive actions such as deleting a project.
This is more than a list of endpoints. It tells the agent how to use those endpoints as a coherent video-production workflow.
Why the Skill Exists
General-purpose AI assistants are good at generating ideas and writing scripts, but video production usually breaks into a chain of disconnected steps:
- Ask an AI for an idea.
- Copy the script into a video tool.
- Choose a voice, character, gameplay background, and caption style.
- Wait for a render.
- Return to the editor to add an image, sticker, or timed callout.
- Export the new version and move it into a publishing workflow.
Every handoff creates another opportunity to lose context or introduce a mistake. A character name may not match a real catalog ID. A script update may never be applied to the project. An overlay may be timed against a guess instead of the rendered caption timeline. A retry may create a duplicate REST job.
BrainrotShorts Agent Skills exist to encode the correct sequence once. The AI assistant can then handle the mechanical coordination while the creator stays focused on the hook, pacing, tone, and final approval.
Agent Skills and MCP Solve Different Parts of the Problem
We have a separate BrainrotShorts MCP setup guide, so there is no need to repeat the protocol details here. The useful distinction is simple:
| Layer | What it gives the agent | BrainrotShorts example |
|---|---|---|
| Agent Skill | The playbook: when to use a tool, in what order, with which checks | Discover catalog IDs, build a draft, review it, request render approval, poll, and report the URL |
| MCP | The live tools available inside an interactive agent conversation | project_create, script_generate, overlay_add, render_start, and related actions |
| REST API | Direct HTTP access for code-driven or server-side automation | Create AI voice-over projects, start renders, and integrate with a custom pipeline |
MCP gives an AI agent hands. The Agent Skill teaches it the job.
The BrainrotShorts skill prefers MCP for interactive chat workflows and uses REST for direct HTTP automation, AI voice-over projects, or environments where MCP is unavailable. It also tells the agent to treat live tool schemas and the canonical API documentation as authoritative, which helps the workflow stay resilient as capabilities evolve.
What Can an AI Assistant Do With the Skill?
Once the skill and a BrainrotShorts connection are available, an agent can help with the full production lifecycle.
Develop ideas and scripts
The assistant can turn a niche, topic, link, or rough idea into a compact creative brief. It can generate hooks, write dialogue variations, adapt a Reddit story, or divide narration into clean production scenes.
The skill encourages a practical short-form structure: lead with the hook, add context after attention is earned, use short spoken beats, build curiosity through the middle, and pay off the opening promise before an optional call to action.
Create and refine video projects
The agent can discover the live BrainrotShorts catalogs before choosing characters, voices, gameplay backgrounds, and caption presets. It can create a draft, apply the selected script, inspect the resulting scenes, and adjust project or scene settings.
This catalog-first behavior matters. The assistant is instructed to use list and get tools instead of inventing IDs that merely sound plausible.
Manage visual assets and timed overlays
For a rendered base video, the agent can read the absolute scene and caption timeline, discover available stickers and fonts, list uploaded images, or upload a remote JPEG, PNG, WebP, or GIF for use as an overlay.
It can then add or update timed text, image, video, and sticker overlays using explicit millisecond timing and normalized canvas positions. That makes requests such as “show the product screenshot when the feature is mentioned” much easier to express conversationally.
Render and retrieve finished videos
When the user explicitly authorizes a finished video, the agent can start the render, poll until it succeeds or fails, and return the latest output URL. The result is a vertical 9:16 MP4 that can move into the next step of the content pipeline.
The approval boundary is deliberate: a draft request stays a draft, while “render,” “export,” or “create the finished video” authorizes the paid render step.
Prepare content for publishing
The BrainrotShorts skill retrieves the final MP4; it does not publish directly to TikTok, Instagram Reels, or YouTube Shorts by itself. In an agent that also has a social publishing connector or scheduling skill, the output URL can become the input to that next workflow.
That composability is one of the strongest ideas behind Agent Skills. One skill produces and verifies the asset; another can prepare the caption, schedule the post, or record the result in a content calendar.
How to Install BrainrotShorts Agent Skills
You need Node.js with npx, a compatible AI agent, and a BrainrotShorts account with access to the integration you plan to use.
Step 1: Install the skill
Open a terminal in the project or working folder used by your AI agent, then run:
npx skills add MaikoCode/brainrotshorts-agent-skills --skill brainrotshorts
The skills CLI downloads the selected skill and configures it for the detected agent. The repository currently lists support for Codex, Claude Code, Cursor, Gemini CLI, GitHub Copilot, OpenCode, and other clients in the open Agent Skills ecosystem.
After installation, start a new agent session if the skill does not appear immediately. You can also inspect the installed SKILL.md before using it; reviewing third-party instructions is a sensible habit for any agent extension.
Step 2: Create the correct BrainrotShorts key
Choose the connection that matches the workflow:
- For interactive agent work, create an MCP-scoped key from the BrainrotShorts MCP Access dashboard.
- For direct HTTP automations, create a separate REST API key from the API dashboard.
MCP and REST keys have different scopes and are not interchangeable. Store either key in your agent's secret configuration or an environment variable. Never paste a real key into a public chat, commit it to Git, or place it in a shared source file.
Step 3: Connect the MCP server for chat-driven workflows
Use the direct endpoint with the www host:
https://www.brainrotshorts.com/api/mcp
A generic client configuration looks like this:
{
"mcpServers": {
"brainrotshorts": {
"url": "https://www.brainrotshorts.com/api/mcp",
"headers": {
"x-api-key": "brs_your_mcp_api_key"
}
}
}
}
Use the client's secret field in place of the placeholder whenever possible. Some clients do not expand environment variables inside JSON, so verify the client-specific behavior instead of assuming.
The www endpoint is important because some clients discard custom authentication headers when following a redirect from the bare domain. The full connection options and tool catalog are in the BrainrotShorts AI agent documentation.
Step 4: Test the connection without rendering
Start with a read-only request:
Use the brainrotshorts skill to list my recent projects and tell me which ones
have a finished render. Do not create, change, or render anything.
If the agent can list the projects, the skill and live BrainrotShorts tools are available. A read-only check also confirms authentication without consuming render credits.
How to Use the Skill With ChatGPT, Claude, Gemini, and Other Agents
The installation command is shared, but invocation syntax and tool configuration vary by client.
ChatGPT and Codex
OpenAI supports skills in the ChatGPT desktop app's Codex experience, Codex CLI, and the Codex IDE extension. In Codex, you can invoke the installed skill explicitly with $brainrotshorts, or describe a matching BrainrotShorts task and let the agent select it.
ChatGPT surfaces that expose Skills or Work mode can select a skill from the composer, commonly with an @ mention. Availability and connection setup can vary by ChatGPT surface and workspace. A normal chat that has neither the installed skill nor access to BrainrotShorts tools cannot render a video simply because the skill exists on GitHub.
Example:
$brainrotshorts Create a 35-second dialogue short about why AI agents need
domain-specific skills. Use two contrasting characters, fast pacing, and bold
captions. Show me the draft and production choices before any render.
Claude Code
Install the skill from the folder where you use Claude Code, ensure the BrainrotShorts MCP server is configured for that environment, and invoke it explicitly or by name.
Use the brainrotshorts skill to turn this Reddit URL into a 50-second story
video. Use Minecraft gameplay, keep identifying details out of the adaptation,
and stop after creating the draft so I can review it.
The skill gives Claude the BrainrotShorts-specific operating procedure; MCP supplies the tools it calls during the conversation.
Gemini CLI
Run the same installer in the Gemini CLI working directory. Then ask Gemini to use the skill and verify that its environment can access either the BrainrotShorts MCP server or the REST API.
Use the installed brainrotshorts skill. Generate three hooks for a short about
beginner investing mistakes, choose the strongest one, and create a draft AI
voice-over project. Do not render yet.
The skill is portable, but Gemini's MCP and secret configuration should follow the current Gemini CLI documentation.
Cursor, GitHub Copilot, OpenCode, and other compatible agents
The same pattern applies to other Agent Skills-compatible tools:
- Install the skill in the agent's project or user skill scope.
- Configure BrainrotShorts MCP or provide the REST key securely.
- Start a fresh session if needed.
- Mention
brainrotshortsexplicitly in the first test prompt. - Keep the first action read-only or draft-only until the connection is verified.
The open Agent Skills specification defines the portable SKILL.md structure. Each client still decides where skills live, how they are invoked, and which external tools are available, so consult that client's documentation when installation discovery or MCP configuration differs.
Practical Workflows You Can Automate
The following examples show where the skill saves the most coordination work.
Workflow 1: Idea to approved dialogue short
Use $brainrotshorts to create a 40-second dialogue video for startup founders.
Topic: users do not want more AI features; they want fewer repetitive steps.
Generate three script variations, recommend the strongest one, and create a
draft with readable captions and a gameplay background. Show me every creative
choice. Render only after I approve the draft.
Behind the scenes, the agent can discover valid characters, backgrounds, and caption styles; create a project; generate script variations; apply the selected script; inspect the scenes; and wait for approval before rendering.
Workflow 2: Reddit URL to narrated short
Use the brainrotshorts skill to adapt this Reddit post into a 55-second video.
Preserve the real conflict and outcome, remove unnecessary usernames, use a
fast narrator and Subway Surfers-style gameplay, and show me the script before
creating the finished video: https://www.reddit.com/...
For speed, the agent can use the one-call Reddit workflow. For editorial control, it can fetch the post, repurpose it, create a draft project, and let you review the story before rendering.
Workflow 3: Add timed visual assets
Use $brainrotshorts on my latest rendered project. Read the caption timeline,
place this product screenshot on screen when the speaker says "automation",
add a reaction sticker at the punchline, and show me the overlay plan before
you re-render: https://example.com/product.png
The agent first reads the rendered timeline, uploads or discovers the assets, applies exact timing, inspects the overlay configuration, and requests approval for the new render.
Workflow 4: Build a batch without losing consistency
Create five draft videos for a weekly AI productivity series. Keep the same
character pair, caption style, background family, 30–45 second duration, and
one-sentence CTA. Give each video a different hook. Return a table with title,
hook, project ID, and draft status. Do not render the batch yet.
This turns a vague batch request into a repeatable production system. The agent can maintain shared creative constraints while tracking each project separately.
Workflow 5: Generate an AI voice-over through REST
For custom automations, the skill can guide an agent through the BrainrotShorts REST API. The core creation call looks like this:
curl "https://www.brainrotshorts.com/api/v1/voiceover/projects" \
-H "X-API-Key: $BRAINROTSHORTS_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: weekly-ai-brief-001" \
-d '{
"title": "Weekly AI brief",
"scenes": [
{"text": "AI agents are moving beyond answers."},
{"text": "Skills give them a repeatable way to finish the work."}
]
}'
The skill reminds the agent to use a unique Idempotency-Key on mutating POST requests, poll the shared render lifecycle, respect rate limits, and record the response requestId when reporting a failure. See the BrainrotShorts REST API reference for the current endpoint schemas.
Workflow 6: Hand the finished asset to a publishing agent
Use $brainrotshorts to retrieve the finished MP4 for the approved project.
Then use my connected publishing workflow to draft platform-specific captions
for TikTok, Reels, and YouTube Shorts. Schedule nothing until I approve the
captions and posting times.
This example requires a separate publishing connector or skill. BrainrotShorts produces the video and exposes the output URL; the agent environment determines whether it can upload or schedule the result elsewhere.
Why This Is One of the Most Direct AI Chat-to-Video Workflows
The advantage is not that every creative decision disappears. It is that the conversation becomes the control surface for the whole production loop.
| Traditional workflow | Agent Skill workflow |
|---|---|
| Copy an idea from chat into a script tool | Develop and revise the idea in the same conversation |
| Manually browse voices, characters, and backgrounds | Ask the agent to inspect the live catalogs and recommend valid options |
| Rebuild settings for every project | Reuse a brief or production pattern across a batch |
| Guess overlay timing in an editor | Place assets against the rendered scene and caption timeline |
| Return later to check a job | Let the agent poll status and return the finished URL |
| Move the export into another workflow manually | Hand the verified output to another compatible agent skill or connector |
This is why BrainrotShorts is an interesting example of AI-powered content creation moving beyond a single “generate” button. The product exposes creation, editing, rendering, asset handling, and retrieval as agent-ready operations, while the skill adds the practical judgment needed to use them well.
The result is a notably short path from “make a video about this” to a reviewable draft or finished MP4—without forcing the creator to bounce between an AI chat, a script document, a video editor, an asset library, and a render-status page for every iteration.
Tips for Better Results
Give the agent enough creative direction to make useful choices without prescribing every setting. A strong brief usually includes:
- Topic and intended audience
- Format: dialogue, Reddit story, or AI voice-over
- Hook or central curiosity gap
- Target duration
- Voice, tone, and pacing
- Preferred characters or narrator, if any
- Background and caption style
- Call to action, or an instruction to omit one
- Whether you want a draft, a render, or a publish-ready handoff
Also separate approval stages in larger workflows. Ask for the script first, then the project draft, then the render, and finally the publishing handoff. Agent automation is most useful when it removes repetitive work without removing the creator's checkpoints.
Frequently Asked Questions
What are BrainrotShorts Agent Skills?
They are reusable, installable instructions that teach compatible AI agents how to operate BrainrotShorts through MCP or its REST API. The skill covers creative briefing, scripts, project creation, assets, overlays, rendering, status checks, and output retrieval.
Do Agent Skills replace MCP?
No. The skill describes the correct workflow; MCP exposes the live BrainrotShorts tools inside an interactive agent. For custom code or environments without MCP, the skill can guide REST API calls instead.
Can ChatGPT create a BrainrotShorts video with the skill?
Yes, when the ChatGPT or Codex surface supports the installed skill and has access to the BrainrotShorts tools. A standard chat without the skill or an MCP/app connection can help write a script, but it cannot operate your BrainrotShorts account or render the video.
Does the skill work with Claude and Gemini?
The repository supports Claude Code and Gemini CLI, along with Codex, Cursor, GitHub Copilot, OpenCode, and other compatible agents. Exact skill discovery, invocation syntax, MCP support, and secret configuration vary by client.
Can the skill publish directly to TikTok, Instagram, or YouTube?
Not on its own. It can return the finished MP4 URL. An agent with a separate publishing or scheduling integration can use that output in a subsequent, approval-gated workflow.
Is the BrainrotShorts Agent Skill only for brainrot dialogue videos?
No. It supports AI brainrot dialogue shorts, Reddit story videos, and AI voice-over projects, plus project management, captions, backgrounds, and timed overlays.
Conclusion
BrainrotShorts Agent Skills turn a capable but general AI assistant into a video-production agent that understands the actual BrainrotShorts workflow. The skill adds repeatable creative and operational guidance; MCP or REST supplies the live actions; and the creator retains control over drafts, renders, and publishing.
For anyone already using ChatGPT, Claude, Gemini, Codex, or another compatible agent, that combination offers a practical way to create brainrot-style short-form videos with fewer handoffs, less configuration repetition, and a much more continuous creative process.