Teaching Skills for Codex

Codex-native packaging of the Teaching Skills suite — the full university teaching lifecycle (design → build → assess → deliver → reflect → improve) for OpenAI Codex CLI. This is the sibling Codex distribution of Teaching Skills for Claude Code.

This repository vendors the suite as a single Codex skill that routes to 15 workflows by intent:

skills/teaching-suite/
  SKILL.md            # router: intent -> workflow, plus ts-* alias emulation
  manifest.json       # adapter metadata + upstream commit pin
  VERSION
  agents/openai.yaml
  ts/                 # vendored content (each upstream SKILL.md -> WORKFLOW.md)
    <15 skills>/WORKFLOW.md + agents/ references/ templates/
    shared/ scripts/ commands/ MODE_REGISTRY.md

Generated, not hand-edited. The canonical source is the Claude Code repo; this package is reproduced by scripts/build_codex.py there, with no drift. Do not edit files here directly — change the upstream suite and regenerate.

Which repo do I want?

  • Claude Code (CLI / VS Code / JetBrains): use YujxZJCN/teaching-skills — native skills, /plugin install, /ts-* slash commands.
  • Codex CLI: use this repo — the single-suite Codex skill with ts-* aliases.

Install or update

Install the skill from this repo with the Codex skill installer:

python3 "$HOME/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py" \
  --repo YujxZJCN/teaching-skills-codex \
  --ref main \
  --path skills/teaching-suite \
  --method git

To update, remove the installed copy and reinstall:

rm -rf "$HOME/.codex/skills/teaching-suite"
python3 "$HOME/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py" \
  --repo YujxZJCN/teaching-skills-codex \
  --ref main \
  --path skills/teaching-suite \
  --method git

On macOS / most Linux, Python 3 is python3. If your only python is Python 3, substitute it.

Use

Ask in natural language ("design a course on machine learning for 90 students"), or use a Claude-style alias as a mode shortcut: ts-course, ts-exam, ts-rubric, ts-feedback, ts-cohort, ts-dashboard, and more. The router (skills/teaching-suite/SKILL.md) dispatches to the right workflow and enforces the suite's non-negotiables (professor checkpoints, no invented context, evidence-bound person-affecting output, integrity-by-design, no student data in the passport).

How this works in Codex. Codex does not register Claude Code slash commands. The ts-* names are emulated aliases: the vendored ts/commands/*.md files are prompt recipes the router reads to pick a workflow and mode — they are not installed commands. Type ts-exam ... (with or without a leading slash) and the router treats it as a mode shortcut. If anything looks wrong after an upstream change, recover by reinstalling (the update step above) — the skill is a regenerated build product, so a clean reinstall always restores a consistent state.

The deterministic validators travel too:

python3 skills/teaching-suite/ts/scripts/check_passport.py course_passport.yaml
python3 skills/teaching-suite/ts/scripts/build_dashboard.py course_passport.yaml

Versioning

Codex package 0.2.0. Vendored from YujxZJCN/teaching-skills@50258caf73fb on 2026-06-15. The package version tracks the adapter independently of the upstream suite version; see manifest.json.

License

MIT — same as the upstream suite. Architecture inspired by academic-research-skills and its Codex sibling.