YouTube Agent Pro for ChatGPT and Codex

24 skills. 11 local Python tools. From video idea to performance review.

Draft scripts, improve titles and thumbnail concepts, plan Shorts, read retention exports and build content calendars from your own material. This OpenAI edition adapts YouTube Agent Pro with native skill metadata, portable packaging, and fixes to analytics edge cases.

Free, MIT-licensed code. No API key or runtime Python dependencies. Your ChatGPT/Codex plan and tool availability still apply. The pack produces drafts, plans and analysis; it includes no YouTube login, uploader, scheduling service or video renderer. It does not promise virality.

Start with Codex

Requires Python 3.9+ for installation and analysis tools:

git clone https://github.com/ahmedsakri/youtube-agent-pro.git
cd youtube-agent-pro/openai
python3 scripts/install.py --dry-run
python3 scripts/install.py

The installer copies all 24 sibling skills to ~/.agents/skills, with shared resources and license notices. It refuses conflicting existing skills. To update intentionally, --force backs up existing folders before replacement. Keep the backups until you verify the update.

For a single project, choose its skill directory:

python3 scripts/install.py --dest /absolute/path/to/your-project/.agents/skills

Use one installation route to avoid duplicate skill names. Start a new Codex conversation; restart your client if the skills do not appear. Try:

Use $yt to plan my next video about beginner photography. My audience uses phones.
Use $yt-script to turn this outline into a 45-second English Short.
Use $yt-analytics to review this export. Tell me what the data cannot establish.

Natural-language requests can also select a relevant skill automatically. See OpenAI's skill guide for invocation and discovery.

Install as a plugin

Current compatible hosts can use plugin.json and .agents/plugins/marketplace.json. On a Codex CLI that provides the plugin commands:

codex plugin marketplace add ahmedsakri/youtube-agent-pro
codex plugin add youtube-agent-pro-openai@youtube-agent-pro

The local marketplace is also available to supported desktop clients when the repository is opened as a project. Select the plugin from its local source. If unavailable, use the standalone installer. This does not publish a plugin to OpenAI's public directory or to a ChatGPT workspace. See plugin packaging.

Start with ChatGPT

From the latest release, download ChatGPT-INSTRUCTIONS.md, ChatGPT-WORKFLOWS.md and the optional ChatGPT ZIP. Paste the instructions into a conversation or project, attach the workflows and your source material, and ask for a concrete result. For computed scores, a Python-capable mode must extract and run the ZIP's tools. Otherwise request a qualitative review. Native installed plugins and skills can be selected with @ where available.

See ChatGPT setup for both routes and their capabilities. Build the same files locally with:

python3 scripts/build_release.py

Workflows

The eleven helpers are ideascore.py, hookscore.py, title.py, thumblint.py, funnel.py, retention.py, deadair.py, chapters.py, swipe.py, ratecard.py and voiceprint.py. They run locally, use no model API and make no network requests. Examples from the openai/ directory:

python3 skills/yt-script/hookscore.py --hook "I tested three phone cameras in the same dark room" --json
python3 skills/yt-package/title.py --title "Three Phone Cameras, One Dark Room" --thumb "WHICH WINS?" --json
python3 skills/yt-analytics/funnel.py --impressions 42000 --ctr 3.1 --avd 2:10 --length 9:40 --json
python3 skills/yt-voice/voiceprint.py /absolute/path/to/transcript.txt --json

Installed skills resolve helpers from their own location. Keep all siblings installed because some share the hook library and operating guide.

Your channel context

Use a supplied voice profile or the conversation. The optional local default is .youtube-agent/voice.md; start from the template, or ask $yt-voice. A missing profile does not block other tasks. Private profiles and exports do not belong in this public repository; .youtube-agent/ is ignored by Git.

Heuristic scores are editing aids, not forecasts. Research claims need sources. Shorts-feed performance must not be diagnosed using long-form thumbnail CTR alone. Sponsorship figures are illustrative USD assumptions unless replaced with supported deal data. Clip selection does not establish copyright permission for source footage.

Development and verification

python3 -m venv .venv
.venv/bin/python -m pip install -r requirements-dev.txt
.venv/bin/python scripts/validate.py
.venv/bin/python -m unittest discover -s tests -v
.venv/bin/python scripts/build_release.py

On Windows, use .venv\Scripts\python.exe instead. Runtime tools and installers use only Python's standard library; PyYAML is solely for development-time metadata validation. See the port analysis and verification report. CI validates, tests and builds on Python 3.9 and 3.12.

License

MIT. Preserve copyright and permission notices in redistributions. Maintained by Ahmed Sakri. This is an independent community project, not an official OpenAI product.