Agent Skills for Codex.
This repository contains personal Codex skills. Standalone skills live in a top-level skill-name/ directory. Related skill systems may live under a grouping directory such as software-design/skill-name/. Every skill directory still contains a required SKILL.md and recommended agents/openai.yaml.
Use /Users/keisuke-yamasaki/.codex/skills/.system/skill-creator/SKILL.md before creating or updating any skill. New skills should be initialized with Skill-Creator's init_skill.py, edited to remove scaffolding, then validated with quick_validate.py.
The Software Design Skill System is grouped under software-design/ and is organized as a router plus implicit skills and routed specialists.
If a Codex harness only scans direct children of a configured skills directory, configure software-design/ as an additional skills root or flatten it during installation. The grouped layout is for repository organization; each nested software-design/<skill-name>/ directory remains a normal skill directory.
Implicit Software Design skills:
software-design-routerpre-implementation-design-reviewcodebase-design-auditarchitecture-drift-reviewdomain-building-blocksaggregate-designbounded-context-designarchitecture-style-decisionclean-architecture-reviewpackage-module-designerror-handling-designapi-interface-designrefactoring-plannercqrs-decisionevent-sourcing-decisionadr-writerdesign-memo-writersoftware-design-skill-forge
Routed or explicit specialist skills set policy.allow_implicit_invocation: false in agents/openai.yaml. Start with software-design/software-design-router/references/skill-map.md when choosing among them.
swift-design-practicesvalue-object-designpolymorphic-branch-refactoringpolishing-documents
These are intentionally left independent. Software Design skills may reference them, but should not duplicate their detailed guidance.
Run the Skill-Creator validator for each skill:
find . -name SKILL.md -exec dirname {} \; | while read -r skill; do
python3 /Users/keisuke-yamasaki/.codex/skills/.system/skill-creator/scripts/quick_validate.py "$skill"
doneIf python3 cannot import yaml, use a temporary venv with PyYAML. Do not vendor validation dependencies into this repo.
Check for scaffold leftovers:
rg -n '(TO)(DO)|\[(TO)(DO)|(template) (placeholder)' . --glob '!swift-design-practices/scripts/swift_design_scan.py'Check generated metadata:
find . -path "*/agents/openai.yaml" -print -exec rg -n "default_prompt|allow_implicit_invocation|short_description" {} \;- Keep
SKILL.mdconcise and procedural. - Put trigger conditions in frontmatter
description, not only in the body. - Keep
SKILL.mdfrontmatter tonameanddescription. - Move long principles, examples, and background into
references/. - Put reusable output forms in
templates/. - Add
evals/trigger-evals.json,evals/output-evals.json, andevals/regression-cases.mdfor trigger and output boundaries. - Preserve
$skill-nameliterally inagents/openai.yamldefault prompts.