Use REPO skills in .codex/skills instead of project command files.
Project skills for previous command workflows:
book-serve: local Jupyter Book servergit-commit: change analysis and commit workflowgit-pr: pull request workflow
When request intent matches below topics, use .claude/skills/* as the operational source:
manim-video-pipeline: scene design/script/render/audio mux/full concatpip-install: install package in.venvand syncrequirements.txtskill-creator: create/update project skill packagedeep-research: 심층 조사, deep research, 종합 분석, in-depth report → outputs/ 에 저장literature-review: 문헌 조사, lit review, 논문 survey, state of the artpeer-review: 동료 검토 시뮬레이션, 논문/노트북/분석 결과 리뷰source-comparison: 여러 소스·방법론·도구 비교 매트릭스
How to execute:
- Read the target
SKILL.md. - Resolve relative paths from the skill directory first.
- Prefer bundled scripts in
scripts/over re-implementing logic. - Load only required files in
references/.
Priority:
- In this repository, REPO scope skills (
.codex/skills/*) take precedence over USER/ADMIN/SYSTEM skills when names overlap.
- Treat
.codex/skills/*and.claude/skills/*as the default operating procedures for recurring tasks. - Keep dependency scopes separated:
requirements.txtfor full local book + video workrequirements-book.txtfor book, Binder, and deploy runtimebinder/apt.txtfor Binder system packages
- When editing notebooks under
book/, use Binder-safe paths and do not assume the current working directory is the notebook folder. - Do not commit local machine artifacts such as
.env, local settings, generated caches, or render outputs unless the user explicitly asks for that. - Prefer small, task-coherent commits that include related config/dependency updates together.