codex2course

Claude Code skills for course creation — from design to narrated lecture video:

Skill	What it does
ai-tutorials	Design AI course syllabus, lectures, and hands-on projects
codex2course	Topic / outline → handout → slide images → PDF
pdf2video	Slide deck → per-slide narration → TTS audio → mp4
makecourse	Orchestration skill for agents — chains the three skills above end-to-end via codex exec

Each skill can be installed and used independently. pdf2video assumes the output layout produced by codex2course. makecourse is intended for automation agents (e.g. openclaw, hermes) that drive the full pipeline by shelling out to the codex CLI.

---

Install

Requires Node.js. Skills are installed to ~/.claude/skills/. Restart Claude Code after installing.

Single skill

npx skills add likefallwind/courseskills --skill ai-tutorials
npx skills add likefallwind/courseskills --skill codex2course
npx skills add likefallwind/courseskills --skill pdf2video
npx skills add likefallwind/courseskills --skill makecourse

All skills

npx skills add likefallwind/courseskills

Manual install (no Node.js)

# ai-tutorials
curl -fsSL https://raw.githubusercontent.com/likefallwind/courseskills/main/skills/ai-tutorials/SKILL.md \
  -o ~/.claude/skills/ai-tutorials.md

# codex2course
curl -fsSL https://raw.githubusercontent.com/likefallwind/courseskills/main/skills/codex2course/SKILL.md \
  -o ~/.claude/skills/codex2course.md

# pdf2video
curl -fsSL https://raw.githubusercontent.com/likefallwind/courseskills/main/skills/pdf2video/SKILL.md \
  -o ~/.claude/skills/pdf2video.md

# makecourse
curl -fsSL https://raw.githubusercontent.com/likefallwind/courseskills/main/skills/makecourse/SKILL.md \
  -o ~/.claude/skills/makecourse.md

---

Prerequisites

codex2course

• Claude Code with imagegen skill installed (used to render slide images)
• Python 3 + Pillow for PDF assembly:

  pip install Pillow

pdf2video

• Everything above, plus:
• ffmpeg on PATH (video assembly)
• MINIMAX account + API key (TTS)
• Python 3 + requests:

  pip install requests

• Set your key before running:

  export MINIMAX_API_KEY=your_key_here

makecourse

• Everything above (the inner skills must be installed and reachable to codex).
• Codex CLI on PATH — the skill drives the pipeline by shelling out to codex exec.
• Run from an agent runtime that can execute shell commands (openclaw, hermes, scheduled bots, etc.). It is not designed for direct human invocation — call the inner skills directly instead.

---

Usage

Start a conversation in Claude Code and describe your goal. The skills trigger automatically from natural language:

# ai-tutorials
Design a 10-lesson LLM application development course for CS undergrads.

# codex2course
Create a 6-hour Python async course for backend engineers.

# pdf2video  
Turn the course in ./course/ into a narrated lecture video, voice: male-qn-qingse.

# Full pipeline
Design a Vibe Coding course, then build slides and produce a narrated mp4.

All four skills are incremental — they inspect what already exists and pick up at the next missing stage, so you can stop, review, and resume without redoing approved work.

Agent-driven full pipeline (makecourse)

When an automation agent (openclaw, hermes, …) needs to produce a full course unattended, it should load the makecourse skill and shell out to the Codex CLI per stage:

codex exec --cd ./course --sandbox workspace-write \
  "Use the ai-tutorials skill to generate a complete course on '<topic>' …"

codex exec --cd ./course/lesson1 --sandbox workspace-write \
  "Use the codex2course skill on this directory …"

codex exec --cd ./course/lesson1 --sandbox workspace-write \
  "Use the pdf2video skill on this directory. TTS provider: edge …"

The skill itself documents the exact prompts, per-lesson loop, and verification checks. See skills/makecourse/SKILL.md.

---

Output layout

course/
├── outline.md          # Course info + image-gen settings
├── handout.md          # Full teaching notes (source of truth)
├── slide-units/        # Per-slide content, derived from handout.md
├── slides/             # Rendered .png pages (000-cover … zzz-ending)
├── course-deck.pdf     # Assembled slide deck
├── audio.md            # TTS voice / padding settings  (pdf2video)
├── narration/          # Per-slide spoken text          (pdf2video)
├── audio/              # Per-slide .mp3 files           (pdf2video)
└── course-video.mp4    # Final narrated lecture video   (pdf2video)

---

Scripts

Both skills ship helper scripts used internally — you can also run them directly:

Script	Purpose
skills/codex2course/scripts/split_handout.py	Slice handout.md into slide-units/
skills/codex2course/scripts/images2pdf.py	Combine slides/*.png into a PDF
skills/pdf2video/scripts/synth_audio.py	Call MINIMAX TTS for each narration file
skills/pdf2video/scripts/assemble_video.py	Pair slides + audio into course-video.mp4

python skills/codex2course/scripts/split_handout.py course/handout.md
python skills/codex2course/scripts/images2pdf.py course/slides course/course-deck.pdf
python skills/pdf2video/scripts/synth_audio.py course/
python skills/pdf2video/scripts/assemble_video.py course/