Curated Codex Skills

A curated, production-tested collection of reusable Codex skills and workflows. Each skill is selected for a concrete workflow, kept auditable to its sources, and exercised through the platforms on which its behavior differs.

Preview: v0.1.0 established the maintenance contract, v0.1.1 hardened upgrades and compatibility checks, and v0.1.2 defaults native controls to English. Interfaces may evolve before v1.0.0; changes are tracked in the changelog.

Catalog

Skill Purpose Origin
prompt-review-and-dispatch Clarify, polish, approve, and continue a prompt in the current conversation by default; explicitly requested background tasks remain supported. Original composition
prompt-master-gpt5 Audit or produce a lean GPT-5.6 prompt. Distilled from nidhinjs/prompt-master plus OpenAI guidance
grill-with-docs Stress-test plans while maintaining domain language and decisions. Adapted from mattpocock/skills
grilling Resolve decisions one native Codex question at a time. Adapted from mattpocock/skills
domain-modeling Maintain project language and durable decisions. Adapted from mattpocock/skills

SOURCES.md pins every upstream commit, source path, hash, and adaptation. AUTHORING-AUDIT.md records the authoring standards used. MIT notices are preserved in THIRD_PARTY_NOTICES.md.

Compatibility

The installers target Codex's documented user skill directory, $HOME/.agents/skills (%USERPROFILE%\.agents\skills on Windows).

The repository exercises installation and validation on GitHub-hosted Linux, macOS, and Windows runners with Python 3.11. A separate Ubuntu matrix runs the validator and tests on Python 3.10 through 3.14. Windows CI executes the shell installer through Git Bash and the PowerShell installer through both Windows PowerShell 5.1 and PowerShell 7. WSL has no retained runner artifact and is not an independently auditable compatibility claim. Native approval additionally depends on the Codex host exposing request_user_input; Windows setup is documented in docs/windows-native-input.md. Agent-authored native controls default to English while normal assistant prose remains language-adaptive. The client controls localization of its built-in Other choice. The evidence and limits for each surface are recorded in the compatibility matrix.

The skills require an interactive, authenticated Codex surface. They do not bundle credentials, enable experimental features silently, or claim support for non-Codex agents.

Install

macOS, Linux, WSL, or Git Bash

git clone https://github.com/George930502/curated-codex-skills.git
cd curated-codex-skills
bash scripts/install.sh

Windows PowerShell

git clone https://github.com/George930502/curated-codex-skills.git
Set-Location .\curated-codex-skills
powershell -NoProfile -ExecutionPolicy Bypass -File .\scripts\install.ps1

The installers copy all catalog skills and report whether default_mode_request_user_input is available and enabled. They do not change Codex configuration. Reinstalling replaces catalog-owned skill directories so deleted files do not linger; unrelated skill directories are preserved. Local edits inside a catalog-named directory are replaced, so keep custom work in a separately named skill. Codex detects skill changes automatically, but a restart may be needed if they do not appear. To test an install without touching your real home, use:

SKILLS_INSTALL_DIR="$(mktemp -d)/skills" bash scripts/install.sh

$target = Join-Path ([System.IO.Path]::GetTempPath()) ([guid]::NewGuid())
powershell -NoProfile -ExecutionPolicy Bypass -File .\scripts\install.ps1 -Destination $target

Validate

Validation is self-contained and uses only the Python standard library:

python3 scripts/validate.py --skills-dir skills
python3 scripts/check_repository.py
python3 -m unittest discover -s tests -v

On Windows, replace python3 with py -3. These are the same repository checks run by CI. Python 3.10 is the minimum supported version. Installer tests always use temporary destinations; see CONTRIBUTING.md for the full contribution gate.

Project contract

  • Contributing — change flow, skill admission, and checks
  • Security — private vulnerability reporting and scope
  • Support — questions, defects, and native-input diagnostics
  • Governance — maintainership and decision policy
  • Roadmap — addition criteria and near-term direction
  • Releases — versioning and release procedure
  • Code of Conduct

This project is licensed under the MIT License.