A collection of independently installable Codex skills for scientific code, manuscript peer review, plotting, web design, and task execution. Each skill defines a focused workflow and carries its own instructions and supporting resources.
This repository is the source of truth for the collection. Start with the catalogue below, install the skills relevant to your work, and use each skill's SKILL.md for its full scope and conventions.
| Skill | Use it for | Scope |
|---|---|---|
| angze-code-style | Reviewing, documenting, pruning, and improving the readability of scientific code and Jupyter notebooks | Preserve calculations, data and interface contracts, and notebook execution state; ground scientific explanations in available evidence. |
| angze-peer-reviewer | Configurable scientific manuscript review, submission stress testing, revision audits, and panel simulation | Experimental v0.2.0 calibration; grounded, fair concerns with separate diagnostic scope, revision budget, and execution authorization. Limited real-manuscript calibration; A/B testing deferred and reviewer realism unvalidated. |
| angze-plot-style | Creating, restyling, reviewing, and exporting Matplotlib scientific figures | Apply Angze's plotting conventions to ordinary figures, manuscript outputs, and compact diagnostics using self-contained plotting code. |
| angze-task-grader | Choosing an appropriate level of effort before a codebase or scientific-computing task | Match inspection, implementation, validation, and review to the task's scope and risk. |
| angze-web-style | Building and refining research sites, technical documentation, portfolios, and writing archives | Apply Angze's web design conventions, including responsive layouts and Day/Night themes, while respecting the site's purpose and structure. |
Skills can be used individually or together. For example, angze-task-grader can guide how much work a task needs while angze-code-style governs preservation and documentation. Each skill's instructions define its boundaries; the collection does not require installing every skill.
Ask Codex to install a skill from its GitHub directory. For example:
$skill-installer
Install https://github.com/angzeli/codex-skills/tree/main/skills/angze-plot-style
Replace angze-plot-style with any name from the catalogue. Install the individual skills/<name>/ directory. The repository root contains collection documentation and tooling and is not itself a skill.
The installed skill should be available on your next turn. If it does not appear, start a new task or restart Codex. For a more detailed walkthrough using angze-code-style, see Getting started.
Use this option when you want runtime skills to follow the contents of a local checkout. It requires Git and a Bash-compatible shell.
git clone https://github.com/angzeli/codex-skills.git
cd codex-skills
./scripts/list_skills.sh
# Preview and install one skill.
./scripts/install_skill.sh --dry-run angze-plot-style
./scripts/install_skill.sh angze-plot-styleTo install the whole collection:
./scripts/install_all_skills.sh --dry-run
./scripts/install_all_skills.shThe shared installers validate each skill and create one symbolic link per skill under $HOME/.agents/skills/. They locate the repository dynamically. An existing conflicting installation is preserved as a timestamped backup before the link is created. Set CODEX_SKILLS_INSTALL_ROOT to use a different installation directory.
Keep the clone in place: the links refer to its skill directories, and updates to those directories become the runtime contents. Start a new Codex task after the first installation.
Name a skill explicitly with $<skill-name> and describe the task, relevant files, and constraints. Codex may also select a skill when the request matches its description.
Scientific code review
$angze-code-style
Review this notebook's documentation and scientific contracts. Do not modify or execute it.
Scientific plotting
$angze-plot-style
Create a manuscript figure from this table. Label the axes with units and export PDF and PNG versions.
Task execution
$angze-task-grader
Fix this parser bug. Choose inspection and validation appropriate to the risk and keep the change focused.
Web design
$angze-web-style
Update this research website to match Angze's design conventions, with responsive layouts and Day/Night themes.
Provide the data, files, or repository context each request needs. State review-only instructions, protected outputs, and other constraints explicitly. Review generated work before accepting it, especially where scientific meaning or stored analysis state is involved.
skills/<name>/ Independently installable runtime package
SKILL.md Entry point, scope, and workflow
... Supporting resources required by that skill
evals/<name>/ Matching prompts, fixtures, rubrics, and evaluation records
scripts/ Shared discovery, validation, fixture, and installation tools
docs/ Evaluation summaries and release documentation
GETTING_STARTED.md Installation walkthrough using angze-code-style
CONTRIBUTING.md Contribution workflow and required checks
CHANGELOG.md Recorded changes across the collection
Runtime resources stay with their skill. Evaluation material and collection tooling stay outside runtime packages. See the runtime directory conventions and evaluation guide for details.
Run shared commands from the repository root:
# Validate all skill packages, including frontmatter and local references.
./scripts/check_all_skills.sh
# Validate one package and run its available fixture checks.
./scripts/check_skill.sh angze-plot-style
./scripts/run_fixture_checks.sh angze-plot-style
# Run all available fixture suites.
./scripts/run_fixture_checks.shReplace angze-plot-style with the skill being changed. Structural checks establish package consistency; they do not establish usefulness or scientific correctness. Fixture checks and recorded evaluations provide evidence only for the cases and runtime versions they cover.
Evaluation material is maintained separately for code style, peer review, plot style, task grading, and web style. Consult evaluation summaries, release records, and the changelog for version-specific results and limitations.
To add a skill, create skills/<name>/SKILL.md with a kebab-case directory name and matching frontmatter name. Include only resources needed at runtime, place evaluation material under evals/<name>/, and reuse the collection's shared scripts. Update this catalogue and the changelog as part of the contribution.
Use synthetic, public-safe fixtures and examples. Check changes for credentials, personal paths, private research, and publication rights before sharing them. Follow CONTRIBUTING.md for the complete workflow and required validation.
This collection is available under the MIT License.
Angze Li