Agent Skills

OpenAI Codex Agent Skills

Codex workflows that know what done means.

Production-oriented instructions, focused skills, and specialist agents for OpenAI Codex. They keep work bounded, ground decisions in the real project, and tie completion to checks that actually ran.

Install the catalog Browse the repository

Operating method

Less ceremony. More proof.

The catalog avoids a universal process for every request. Each skill has a narrow trigger, loads deeper guidance only when needed, and keeps completion tied to observable evidence.

01

Evidence before edits

Repository rules, nearby patterns, versions, and runtime behavior come before assumptions. Current external facts are checked against primary sources.

02

Smallest complete change

The agent solves the requested problem without speculative abstractions, unrelated cleanup, fake implementations, or hidden scope expansion.

03

Risk-based verification

Checks start close to the changed behavior and expand only when the risk justifies it. A polished diff is not treated as proof.

04

Human reports

Ordinary answers use connected paragraphs, not note-like bullet lists. Engineering reports explain the outcome and approach in two short paragraphs, then place changes, commands, checks, and sources in compact tables.

Focused catalog

One skill for the job at hand.

Thirteen focused Codex skills cover specification, architecture, implementation, review, research, interface design, and skill governance without turning every request into the same workflow.

Specify and plan

Clarify only decisions that evidence cannot resolve. Build a task plan when planning is requested, without diverting an already authorized implementation.

  • clarify-and-specify
  • spec-to-tasks

Model and structure

Clarify business meaning and design module, data, runtime, deployment, and team boundaries from measurable quality and change drivers.

  • design-project-architecture
  • domain-modeling

Build and verify

Separate causes from symptoms, implement approved behavior, use test-first work only when required, and review concrete changes against relevant production risks.

  • diagnose-root-cause
  • production-code-quality
  • tdd
  • implement-and-verify
  • review-changes

Research and govern

Verify current claims, audit reusable skills, preserve evidence-backed workflows, and design interfaces around user tasks rather than generated decoration.

  • research-current-sources
  • audit-agent-skills
  • learn-from-history
  • design-ui-ux

Codex design-ui-ux

Designed for a task, not generated for a screenshot.

The UI/UX skill scales from a single setting to a governed design system. It starts with user evidence and platform behavior, preserves accessibility and search visibility where they apply, and rejects decoration without a product reason.

Clear hierarchy. Real states. Decisions with an owner.

Evidence comes before style

Product needs and real content lead. Platform guidance and the local design system then constrain layout, interaction, language, and visual expression.

Complexity changes the process

A local view needs a clear task and feedback. A complex product also needs roles, permissions, data responsibility, recovery, and complete state coverage.

A design system needs governance

Tokens and components need an owner, evidence, review, release notes, usage guidance, and a migration path. A component library alone is not a system.

Generated UI remains untrusted

Real data, semantics, states, accessibility, runtime behavior, and human review decide release readiness. A polished screenshot proves none of them.

Installation

Keep your setup or replace it.

Choose merge to preserve your current configuration. Choose replace to install the complete opinionated setup. Both modes back up every file they change.

Merge with your setup

Keep existing Codex settings, add the agents and skills, and connect the shared operating instructions.

cp -R "$HOME/.codex" "$HOME/.codex.backup-$(date +%Y%m%d-%H%M%S)" 2>/dev/null || true
git clone https://github.com/dayfinggg/openai-codex-agent-skills.git
mkdir -p "$HOME/.codex/agents" "$HOME/.codex/skills"
cp -R openai-codex-agent-skills/codex/agents/. "$HOME/.codex/agents/"
cp -R openai-codex-agent-skills/codex/skills/. "$HOME/.codex/skills/"
cp openai-codex-agent-skills/codex/model-instructions.md "$HOME/.codex/"

Replace everything

Back up your current directory, then replace the managed Codex configuration with the complete setup.

cp -R "$HOME/.codex" "$HOME/.codex.backup-$(date +%Y%m%d-%H%M%S)" 2>/dev/null || true
git clone https://github.com/dayfinggg/openai-codex-agent-skills.git
rm -rf "$HOME/.codex/agents" "$HOME/.codex/skills"
mkdir -p "$HOME/.codex"
cp -R openai-codex-agent-skills/codex/agents \
  openai-codex-agent-skills/codex/skills "$HOME/.codex/"
cp openai-codex-agent-skills/codex/model-instructions.md \
  openai-codex-agent-skills/codex/config.toml "$HOME/.codex/"

For merge mode, add model_instructions_file = "model-instructions.md" as a top-level setting in config.toml. Windows PowerShell commands are in the README. Restart Codex after installation.