by beyondtahir
Turn long GTA 6 streams into viral 9:16 reels automatically. Free, local, works in Claude and ChatGPT.
Available after the catalog security scan passes.
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM lists this project for discovery and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
# Add to your Claude Code skills
git clone https://github.com/beyondtahir/beyondgtaclipsGuides for using ai agents skills like beyondgtaclips.
Long streams in, clean viral clips out. You run everything; the user only says what they want.
SKILL_DIR = the folder this file is in. Every command is:
python3 "SKILL_DIR/scripts/beyondgtaclips.py" <command> ...python "SKILL_DIR\scripts\beyondgtaclips.py" <command> ... (or py)Below, bgc is short for that. Clips are saved in ~/BeyondGTAClips/<date>/<topic>/<creator>__<title>/.
Run long jobs in the background and tell the user in plain words what is happening
(searching, downloading audio, listening, picking moments, making clips).
bgc setup and wait for READY (private Python env in ~/.beyondgtaclips, Whisper
models ~220 MB, face model). Needs ffmpeg and deno or node:
macOS brew install ffmpeg deno · Windows winget install Gyan.FFmpeg and winget install DenoLand.Deno.
Ask the user before anything that needs an admin password.bgc doctor shows what is installed and which AI will pick moments.The user may know nothing about clipping. Guide them in plain words, one step at a time, and tell them what is happening while long steps run. Never show them raw commands unless they ask.
Step 0: first time only. If bgc doctor says it is not set up, say "I'll set things up
first, it takes a few minutes once" and run bgc setup (section 0).
Step 1: ask three questions in ONE short message (skip any the user already answered):
Let's make your GTA 6 clips. Three quick questions:
- How many clips do you want? (default 4)
- How recent should the streams be: last 24 hours, 48 hours or 1 week? (default 48 hours)
- Which style? Here are the five. Pick a number, several (like 1,3), or all. If you don't mind, I'll use 1, Hook card.
Show the style picture with question 3: SKILL_DIR/assets/styles.png. If your app can show
images in the chat, show it there; otherwise open it on the user's screen
(macOS open "<path>", Windows start "" "<path>") and also list the names:
1 Hook card · 2 Stack · 3 Bubble · 4 Cinematic · 5 Breaking news (bgc styles prints this).
If the user says "default", "you choose" or "just do it": 4 clips, 48 hours, style 1. Then repeat their choices back in one line ("4 clips, last 48 hours, Hook card style: starting now") and go on.
Step 2: find the most viewed streams.
bgc find --within 48h --count 6 (topic defaults to GTA 6; --topic "Fortnite" for anything else).
It runs ~13 searches with every spelling (GTA 6, GTA VI, GTA6), drops compilations,
re-uploads, trailers, reviews, shorts and streams already clipped, checks the spoken
language (English first) and ranks finished streams by most views in the window.
Streams that are still live are listed separately (they can be clipped after they end).
Show the top streams in a short list (creator, title, views, how old) and ask "Which ones
should I clip? (or I take the top ones)". If they said "just do it", take the top
ceil(clips / 3) streams yourself and say which.
Step 3: clip. Run it in the background and tell the user what is happening.
bgc clip --pick 1,3 --total 4 --styles 1 (stream numbers from the last find; total clips
and styles from step 1). Other forms: bgc clip URL [URL ...] --total 4 --styles 1,3, or in
one go bgc clip --find --within 24h --total 6 --styles all.
A run takes ~10-20 minutes per stream. While it runs, give the user a short update at each
stage: looking at the stream, listening, picking moments, making clips.
What it does for each stream:
--length long (60-90 s story clips), --focus "police chases",
--section 1:30:00-3:00:00 (only that part), --brain agent|none (see section 3).Step 4: look at every clip and fix it (this is what makes it smart).
Each clip has a check image: <folder>/checks/<clip>__<style>.jpg (3 frames). Open and
look at them. Does the hook describe what is on screen? Is anything cut off or covered?
bgc remake "<folder>" --clip 2 --hook "Speed gets *attacked by an alligator* in the swamps in *GTA 6* 😭"
(*stars* = words shown in bold red; at most one emoji, at the end)bgc remake "<folder>" --clip 2 --styles 3--title, --headline, --caption with remake.Step 5: hand over. Give the user the folder, list each clip with its hook and styles,
and point to POSTING.md (caption, hashtags, two A/B hooks, trending-sound tip per clip).
Every clip is GTA gameplay first: the skill looks at YouTube's preview frames of the whole stream, skips the waiting room, menus and just-chatting parts, and checks each clip's real frames before rendering (a static picture is swapped for a backup moment).
Three styles with text on top and no facecam panel:
| Style | Looks like | Best for |
|---|---|---|
hookcard (1) |
the story sentence on black at the top with the action words in bold red (like the "Speed gets attacked by an alligator in GTA 6" reels), gameplay in the middle, red captions under it, empty band below for the app's buttons | the default viral look: NPC chaos, fails, discoveries, shocking moments |
cinematic (4) |
blurred backdrop, wide gameplay with a slow push-in, letter-spaced label and headline on top, quiet subtitles, fades | beautiful or emotional scenes, slow stories |
news (5) |
BREAKING tag and condensed headline on top, wide gameplay, scrolling red ticker | secrets, hidden features, big discoveries |
Two styles with the streamer's face:
| Style | Looks like | Best for |
|---|---|---|
stack (2) |
the streamer in a white-framed box on top (never stretched), a yellow divider line carrying the title on white leaning boxes, one big leaning caption word, then the whole 16:9 stream screen below | reactions where the streamer's face adds to the moment |
bubble (3) |
full-screen gameplay, the streamer's face in a small round bubble at the bottom, big 2-word pop captions | fast hype moments |
Every style keeps text out of the TikTok/Reels button zones and works in Instagram's 4:5 feed crop.
stack needs a facecam; without one it falls back to hookcard automatically.
--brain)auto uses the CLI of the assistant running this skill, else any installed one, else no AI.
| You are | Use |
|---|---|
| Claude Code | auto (calls claude -p) |
| ChatGPT / Codex | auto = agent mode: you pick the moments yourself from the prompt files (works inside Codex's sandbox). --brain codex calls codex exec instead |
| Kimi Code | --brain kimi |
| Gemini CLI / Cursor / OpenCode | --brain gemini / --brain cursor / --brain opencode |
| MiniMax, ChatGPT app, any other assistant | --brain agent (you pick the moments yourself, see below) |
| No AI at all | --brain none (loudness + chat; never fails to make clips) |
| Any other CLI | --brain custom --brain-cmd "mycli --print" (prompt on stdin) |
Agent mode. The command stops with exit code 3 and writes AGENT_TASK.md next to the
prompt files. Read each .prompt.txt, follow it, write only the JSON array to its
.answer.json, then run the exact same command again. It stops once more for the story
check (STORY_TASK.md): open the contact-sheet images each .story.prompt.txt lists, look
at them, write only the JSON object to its .story.answer.json, and run the same command
again. It continues and makes the clips.
If the AI finds too few strong moments, the loudest moments fill the gap: a stream never ends with zero clips.
~/BeyondGTAClips and
~/.beyondgtaclips, outside the project folder. Codex will ask the user for permission:
tell them "ChatGPT will ask to let me use the internet and save files, please choose
Allow (or Always allow)" before the first command.clip stops with exit code 3, read AGENT_TASK.md, answer each
.prompt.txt with only its JSON array in the matching .answer.json, then run the exact
same clip command again. In agent mode only the 6 hottest parts are prepared, to keep
the reading short.<folder>/checks/*.jpg.~/BeyondGTAClips/settings.json (any key from scripts/bgt/settings.py),
e.g. {"accent": "FF3B3B", "clips": 5, "template": "hookcard"}."clean_captions": "off" shows everything, "all" also
softens swears. The audio is never changed.~/BeyondGTAClips/watchlist.txt (YouTube, Twitch and Kick).bgc prune --days 7 deletes old cached downloads.--cookies chrome (the browser the user is signed in with;
ask the user first, it reads that browser's YouTube login).~/.beyondgtaclips/venv/bin/pip install -U yt-dlp.--brain auto.See how beyondgtaclips compares with popular alternatives.
[!TIP] ⭐ Star this repo so you can find it again when the next big stream drops. It takes one click and it helps other clippers find a free tool.
Clips used to be picked from the sound alone. Sometimes the hook promised something that never showed up, or the cut stopped one second before the punchline. Now your AI looks at the frames of every moment before it cuts:
Works the same in Claude and ChatGPT (and Kimi, Gemini or no AI at all). Nothing new to learn: you ask for clips exactly like before.
MrBeast pays people to clip his videos. So do IShowSpeed, Kai Cenat and Kick. Brands put money into a pot, you post short clips on your own TikTok, Reels or Shorts, and you get paid for every 1,000 views your clips get.
GTA 6 just launched. The biggest streamers are live for 10 hours a day, and millions of people want the 30 seconds where something crazy happens: the alligator attack, the police chase, the NPC meltdown, the streamer losing it.
The clippers who post those 30 seconds first get the views. The hard part was always the same: watching hours of footage, finding the moment, cutting it, making it vertical, writing captions... for every single clip.
That part is now automatic. You answer 3 questions. Your AI assistant finds the most viewed streams, watches them, picks the moments that go viral and hands you finished reels with a hook, captions and a ready-to-paste post.
Real programs and real rates, checked in November 2026. Rates change with every campaign, so always check the live campaign page before you post.
| Who pays | How it works | Typical pay |
|---|---|---|
| 🟣 Content Rewards (paid out through Whop) | The biggest clipping marketplace. Hundreds of campaigns from streamers and brands, including IShowSpeed and Kai Cenat. Join a campaign, post, submit your link. | About $1 per 1,000 views (median), usually capped per clip ¹ |
| 🐐 MrBeast (his own platform, Vyro) | Paid rounds for new MrBeast and MrBeast Gaming videos, plus sponsor rounds. 18+, clips must be approved and pass 5,000 views. | $1-2.50 per 1,000 views, up to $1,000 per clip ² |
| 🟢 Kick streamers | Kick funds clipping campaigns through agencies, and some Kick streamers run their own programs. | One streamer paid $1.4M to 303 clippers in 5 weeks ³ |
| ▶️ YouTube Shorts | Grow a GTA 6 clip channel and join the YouTube Partner Program. You keep 45% of your share of the Shorts ad pool. | A few cents per 1,000 views, so it adds up only at big volume ⁴ |
| 🎵 TikTok Creator Rewards | Pays for qualified views on original videos over 1 minute. Needs 10K followers and 100K views in 30 days. Use long mode. | Depends on country and watch time ⁵ |
| 🤝 Sell it as a service | Streamers pay clippers to grow their channel. Show them reels you made from their stream. | You set the price |
At the typical $1 per 1,000 views campaign rate:
| Your clip gets | You earn |
|---|---|
| 50,000 views | $50 |
| 250,000 views | $250 |
| 1,000,000 views | $1,000 |
| 10 clips × 100,000 views | $1,000 |
[!WARNING] Real talk. These are typical market rates, not promises. Many campaigns only pay for approved clips with verified views, set a minimum view count, cap the pay per clip, and take a platform fee. Budgets can run out in a day. Only clip creators who allow it, and follow each campaign's rules (hashtags, disclosure, length).
| By hand | Beyond GTA Clips | |
|---|---|---|
| Find the biggest streams | Scroll YouTube for an hour | Most viewed from the last 24h / 48h / week, ranked for you |
| Find the moment | Watch 10 hours | Skips menus and waiting, listens for screams, laughs and chat spikes |
| Cut and crop to 9:16 | 20 minutes per clip | Automatic, with the facecam handled |
| Hook + captions | Type them yourself | Word-by-word captions and a scroll-stopping hook |
| Caption + hashtags | Write them yourself | POSTING.md with caption, hashtags and 2 A/B hooks |
| Cost | Your whole day | $0, no API keys, no subscription |
POSTING.md with the caption and hashtags for each one.| Step | What happens behind the scenes |
|---|---|
| 🔎 Finds the streams | Searches YouTube for the most viewed GTA 6 streams in your time window, skips re-uploads and compilations, checks the language. |
| 👀 Watches the whole stream | Looks at the entire stream in seconds and skips waiting rooms, menus and just chatting. Only real gameplay. |
| 👂 Listens for the hype | Finds the moments where the streamer screams, laughs and chat explodes. |
| 🧠 Picks the viral moments | Your AI chooses the funniest, most shocking moments and writes a hook for each. |
| ✂️ Cuts only what it needs | Downloads only those seconds of video (plus a little extra on each side), not the whole 10 hours. |
| 🎞️ Checks the story | Looks at the frames and re-cuts each moment so it shows the whole scene, from just before the surprise to the end of the reaction. Rewrites a hook that doesn't match the picture. |
| 🎬 Edits like a pro | 9:16, hook |
beyondgtaclips is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by beyondtahir. Turn long GTA 6 streams into viral 9:16 reels automatically. Free, local, works in Claude and ChatGPT. It has 51 GitHub stars.
beyondgtaclips's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/beyondtahir/beyondgtaclips" and add it to your Claude Code skills directory (see the Installation section above). beyondgtaclips ships a SKILL.md manifest, so compatible agents can discover and load it automatically.
beyondgtaclips is primarily written in Python. It is open-source under beyondtahir on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh beyondgtaclips against similar tools.
No comments yet. Be the first to share your thoughts!