Back to search
ashtonvossai
MarketplaceWorks with Claude CodeGitHub
Free Claude Code skills from Ashton Voss AI.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/abhishek047--codex-clipper-skill)HTML
<a href="/listings/abhishek047--codex-clipper-skill"><img src="/api/listings/abhishek047--codex-clipper-skill/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Option A: paste this into Claude Code (easiest)
Install the GTA 6 Clipper skills from https://github.com/vossashtonai/gta6-clipper. Clone the repo to a temporary folder, then copy BOTH folders inside its skills/ folder (gta6-clipper and gta6-live-clipper) into ~/.claude/skills/ so they sit side by side. Then run: bash ~/.claude/skills/gta6-clipper/scripts/setup.sh Tell me what it installed and whether it says READY. Then remind me to restart Claude Code.
Option B: as a Claude Code plugin
In Claude Code:
/plugin marketplace add vossashtonai/gta6-clipper /plugin install gta6-clipper@ashtonvossai
Then say **set up the GTA 6 clipper**.
Option C: by hand
git clone https://github.com/vossashtonai/gta6-clipper.git /tmp/gta6-clipper cp -R /tmp/gta6-clipper/skills/gta6-clipper /tmp/gta6-clipper/skills/gta6-live-clipper ~/.claude/skills/ bash ~/.claude/skills/gta6-clipper/scripts/setup.sh
Setup checks what your Mac needs and installs what's missing: Homebrew packages (FFmpeg, whisper-cpp, Deno), a Python environment (yt-dlp, OpenCV, Pillow) and the Whisper speech model (~465 MB). Re-run it any time downloads start failing; it updates yt-dlp.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 15 lines
- {
- "name": "ashtonvossai",
- "owner": { "name": "Ashton Voss AI" },
- "description": "Free Claude Code skills from Ashton Voss AI.",
- "plugins": [
- {
- "name": "gta6-clipper",
- "source": "./",
- "description": "Clip GTA 6 streams into ready-to-post vertical clips, from a finished stream or live. Mac only.",
- "version": "1.0.0",
- "author": { "name": "Ashton Voss AI" },
- "category": "productivity"
- }
- ]
- }
README.md · README · 133 lines · 1 finding marked
- # GTA 6 Clipper
- Two macOS skills for Codex and Claude Code that turn GTA 6 streams into ready-to-post vertical clips.
- - **gta6-clipper** clips a stream that has finished. Give it a YouTube, Twitch or Kick link, or nothing and it finds today's biggest GTA 6 streams.
- - **gta6-live-clipper** clips a stream while it's still live. A clip is ready about 5 to 8 minutes after the moment happens.
- Every clip comes out 1080x1920: facecam on top, gameplay below, a hook title, word-by-word captions, and a post caption with hashtags.
- **Mac version.** On Windows? Use the Windows version: https://github.com/vossashtonai/gta6-clipper-windows
- or paste this in PowerShell: `irm https://raw.githubusercontent.com/vossashtonai/gta6-clipper-windows/main/install.ps1 | iex`
- Note: Runs a downloaded PowerShell script (iex / Invoke-Expression)
- Line 11 on GitHub
- ## Use in Codex
- Open this repository as your Codex workspace. The two links in `.agents/skills/`
- register the existing skill folders with Codex. Start a new Codex chat in this
- workspace if an existing chat does not show them.
- - `$gta6-clipper set up the GTA 6 clipper` installs the shared macOS dependencies. Wait for `READY`.
- - `$gta6-clipper clip this stream <VOD URL>` selects and renders up to three highlights.
- - `$gta6-live-clipper clip this stream live <live URL>` keeps the recorder and highlight selection running in the active chat.
- In Codex, outputs go under the repository's `runs/` folder by default.
- Each clip has an MP4 and a companion text file with a title, caption, and hashtags.
- ## Install in Claude Code
- ### Option A: paste this into Claude Code (easiest)
- ```text
- Install the GTA 6 Clipper skills from https://github.com/vossashtonai/gta6-clipper.
- Clone the repo to a temporary folder, then copy BOTH folders inside its skills/ folder
- (gta6-clipper and gta6-live-clipper) into ~/.claude/skills/ so they sit side by side.
- Then run: bash ~/.claude/skills/gta6-clipper/scripts/setup.sh
- Tell me what it installed and whether it says READY. Then remind me to restart Claude Code.
- ```
- ### Option B: as a Claude Code plugin
- In Claude Code:
- ```text
- /plugin marketplace add vossashtonai/gta6-clipper
- /plugin install gta6-clipper@ashtonvossai
- ```
- Then say **set up the GTA 6 clipper**.
- ### Option C: by hand
- ```bash
- git clone https://github.com/vossashtonai/gta6-clipper.git /tmp/gta6-clipper
- cp -R /tmp/gta6-clipper/skills/gta6-clipper /tmp/gta6-clipper/skills/gta6-live-clipper ~/.claude/skills/
- bash ~/.claude/skills/gta6-clipper/scripts/setup.sh
- ```
- Setup checks what your Mac needs and installs what's missing: Homebrew packages (FFmpeg, whisper-cpp, Deno), a Python environment (yt-dlp, OpenCV, Pillow) and the Whisper speech model (~465 MB). Re-run it any time downloads start failing; it updates yt-dlp.
- ## Use it
- Open Claude Code and say:
- - **clip this stream** `<link>`
- - **find me today's biggest GTA 6 streams and clip them**
- - **live clip** `<link>` (for a stream that's live right now)
- Clips land in `~/GTA6-Clips/`. Each one has an `.mp4` and a `.txt` with the title, caption and hashtags.
- ## Terminal commands
- Run `make help` from the repository root for all setup, VOD, live, thumbnail,
- and publishing commands. Make uses `runs/` for VODs and `runs/live/` for live
- recordings by default. Set `GTA6_CLIPS_DIR` or `GTA6_LIVE_DIR` to use other folders.
- ```sh
- make setup
- make discover ARGS='--limit 5'
- make prep URL='https://www.youtube.com/watch?v=VIDEO_ID'
- make status RUN='runs/example'
- make render RUN='runs/example' ARGS='--top 3'
- make setup-upload
- make upload-preview
- make upload
- ```
- Highlight selection is handled by the clipper skill or by editing pick files
- before running `make render` or `make live-clip`. Pass additional CLI flags
- and multiple positional inputs with `ARGS`. Quote paths individually inside it,
- for example `make thumbnail ARGS='"path/to/first clip.mp4" "path/to/second clip.mp4" --at 1'`.
- ## How it works
- | Step | Finished stream | Live stream | Tools |
- |---|---|---|---|
- | 1. Get the stream | Downloads the audio only (4 h ≈ 220 MB) | Records 1-minute pieces, keeps the last 15 min | yt-dlp, FFmpeg |
- | 2. Speech to text | Whole stream in one go | Each minute as it lands | FFmpeg, Whisper |
- | 3. Find the hype | Loud seconds + YouTube replay chat speed | Loud seconds + live Twitch chat speed | FFmpeg, chat |
- | 4. Pick the moments | Claude reads 12-min chunks, best 3 scoring 7+ | Claude reads each 5-min window, any 7.5+ | Claude Code |
- | 5. Make the clip | Downloads only those minutes | Cuts from the pieces already on your Mac | OpenCV, Whisper, Pillow, FFmpeg |
- Chat is counted, never read: 10-second buckets, a spike is 8+ messages at 2.5x the normal rate. Loud means 9 dB above the streamer's own normal. The facecam is found by looking for a face, and box edges, that stay still while the gameplay moves.
- ## Run it free
- Picking (step 4) is the only part that uses AI. Everything else runs on your Mac. To run without a Claude subscription:
- 1. Install [Ollama](https://ollama.com).
- 2. Check your memory (Apple menu → About This Mac → Memory). 16 GB: `ollama pull qwen3.5:9b`. 32 GB or more: `ollama pull qwen3.5:27b`.
- 3. In Terminal: `ollama launch claude` and pick the model.
- 4. Same prompts as above.
- The free model is slower and its picks are rougher than Claude's. Expect some back and forth.
- ## Good to know
- - Age-restricted YouTube streams and Kick need your browser login. Tell Claude to use `--cookies chrome`.
- - YouTube streams that are still live can be clipped by the finished-stream skill once the replay is up. Twitch past broadcasts work while the stream is still going.
- - Twitch in "find me streams" is optional: create a free app at [dev.twitch.tv/console](https://dev.twitch.tv/console) and put `TWITCH_CLIENT_ID` and `TWITCH_CLIENT_SECRET` in `skills/gta6-clipper/.env` (see `.env.example`).
- - Live clipping needs the Claude Code session to stay open, and run at most 2 streams at once on a 16 GB Mac.
- - Clip streamers who are fine with clipping, and watch every clip before you post it.
- - YouTube ad money needs the Partner Program, and YouTube can refuse it for reposted, mass-made content. Add something of your own.
- ## Made by
- [Ashton Voss AI](https://www.youtube.com/@ashtonvossai). More free AI systems in the AI Alliance Community.
- MIT licensed.
- ## Upload approved clips
- Run `python -m upload.tui` to review runs and publish approved clips to configured
- YouTube, Facebook Page, and Instagram destinations. See the [upload setup guide](upload/README.md).
skills/gta6-clipper/scripts/setup.sh · Bundled script · 80 lines · 1 finding marked
- #!/usr/bin/env bash
- # GTA 6 Clipper setup: checks what this Mac needs, installs what's missing, says when it's ready.
- # Safe to re-run any time.
- set -uo pipefail
- SKILL_DIR="$(cd "$(dirname "$0")/.." && pwd)"
- VENV="$SKILL_DIR/.venv"
- MODELS="$SKILL_DIR/models"
- mkdir -p "$MODELS"
- ok() { printf " [ok] %s\n" "$1"; }
- inst() { printf " [install] %s\n" "$1"; }
- fail() { printf " [MISSING] %s\n" "$1"; FAILED=1; }
- FAILED=0
- echo "GTA 6 Clipper setup"
- if [[ "$(uname)" != "Darwin" ]]; then
- echo "This is the Mac version. On Windows, install the Windows version instead:"
- echo " Open PowerShell (Start -> type PowerShell) and paste:"
- echo " irm https://raw.githubusercontent.com/vossashtonai/gta6-clipper-windows/main/install.ps1 | iex"
- Note: Runs a downloaded PowerShell script (iex / Invoke-Expression)
- Line 19 on GitHub
- echo " (or see https://github.com/vossashtonai/gta6-clipper-windows)"
- exit 1
- fi
- # 1. Homebrew
- if command -v brew >/dev/null 2>&1; then ok "Homebrew"; else
- fail "Homebrew (install from https://brew.sh then re-run setup)"; exit 1; fi
- # 2. Command-line tools: ffmpeg (cutting/encoding), whisper-cpp (local transcript), deno (YouTube downloads)
- for pair in "ffmpeg:ffmpeg" "whisper-cli:whisper-cpp" "deno:deno"; do
- bin="${pair%%:*}"; formula="${pair##*:}"
- if command -v "$bin" >/dev/null 2>&1; then ok "$formula"; else
- inst "$formula"; brew install "$formula" >/dev/null 2>&1 && ok "$formula" || fail "$formula (brew install $formula)"; fi
- done
- # 3. uv (fast Python + venv manager)
- if command -v uv >/dev/null 2>&1 || [[ -x "$HOME/.local/bin/uv" ]]; then ok "uv"; else
- inst "uv"; curl -LsSf https://astral.sh/uv/install.sh | sh >/dev/null 2>&1 && ok "uv" || fail "uv"; fi
- UV="$(command -v uv || echo "$HOME/.local/bin/uv")"
- # 4. Python environment: yt-dlp (always latest, sites change weekly), OpenCV (facecam finder), Pillow (captions)
- if [[ ! -x "$VENV/bin/python" ]]; then inst "Python 3.12 environment"; "$UV" venv --python 3.12 "$VENV" >/dev/null 2>&1; fi
- "$UV" pip install --python "$VENV/bin/python" -q -U "yt-dlp[default]" opencv-python-headless pillow numpy >/dev/null 2>&1 \
- && ok "Python packages (yt-dlp, opencv, pillow, numpy)" || fail "Python packages"
- # 5. Whisper model (speech -> text, runs on this Mac). Reuse an existing one if present.
- WHISPER_MODEL=""
- for m in "$MODELS/ggml-small.en.bin" "$MODELS/ggml-small.bin" "$HOME/.cache/whisper-cpp/ggml-small.en.bin" "$HOME/.cache/whisper-cpp/ggml-small.bin"; do
- [[ -f "$m" ]] && WHISPER_MODEL="$m" && break; done
- if [[ -z "$WHISPER_MODEL" ]]; then
- inst "Whisper small.en model (~465 MB)"
- curl -L -s -o "$MODELS/ggml-small.en.bin" https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-small.en.bin \
- && WHISPER_MODEL="$MODELS/ggml-small.en.bin" || fail "whisper model"
- fi
- [[ -n "$WHISPER_MODEL" ]] && ok "Whisper model: $WHISPER_MODEL"
- # 6. Face detector model (YuNet, ~230 KB) used to find the facecam box
- YUNET="$MODELS/face_detection_yunet_2023mar.onnx"
- if [[ ! -f "$YUNET" ]]; then
- inst "YuNet face detector"
- curl -L -s -o "$YUNET" https://github.com/opencv/opencv_zoo/raw/main/models/face_detection_yunet/face_detection_yunet_2023mar.onnx || fail "YuNet model"
- fi
- [[ -s "$YUNET" ]] && ok "Face detector" || fail "Face detector"
- # 7. Fonts (macOS ships these)
- [[ -f "/System/Library/Fonts/Supplemental/Arial Black.ttf" ]] && ok "Caption font (Arial Black)" || fail "Arial Black font"
- # 8. Free space check: audio-only download keeps this small, but warn early
- FREE_GB=$(df -g "$SKILL_DIR" | awk 'NR==2{print $4}')
- if (( FREE_GB < 3 )); then echo " [warn] Only ${FREE_GB} GB free. Each stream needs ~0.5-1 GB while it runs."; else ok "Disk space (${FREE_GB} GB free)"; fi
- cat > "$SKILL_DIR/config.json" <<EOF
- {"whisper_model": "$WHISPER_MODEL", "yunet_model": "$YUNET"}
- EOF
- if [[ $FAILED -eq 0 ]]; then
- date > "$SKILL_DIR/.ready"
- echo; echo "READY. Say: clip this stream <link> or find me today's biggest GTA 6 streams and clip them"
- else
- rm -f "$SKILL_DIR/.ready"; echo; echo "NOT READY: fix the [MISSING] items above and re-run setup."; exit 1
- fi
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.