Codex Skill

Priorities

Correctness > Security > Efficiency

Goal

Execute OpenAI Codex CLI for automated code analysis, refactoring, and editing tasks. Default to gpt-5.3-codex model with user-specified reasoning effort. Suppress stderr thinking tokens by default unless debugging is needed.

Constraints

Default model: gpt-5.3-codex (ask user for reasoning effort: high, medium, or low)
Sandbox mode: --sandbox read-only (default), workspace-write (for edits), danger-full-access (network/broad access)
Always use --skip-git-repo-check flag
Suppress stderr by default: append 2>/dev/null to all codex exec commands
Resume sessions: echo "prompt" | codex exec --skip-git-repo-check resume --last 2>/dev/null (no config flags between exec and resume unless user specifies)
Ask permission before using high-impact flags (--full-auto, --sandbox danger-full-access)
Stop and report on non-zero exit codes
Inform user after completion: "You can resume this Codex session at any time by saying 'codex resume'"

Model Options

Model	Best for	Context window	Key features
`gpt-5.3-codex` ⭐⭐	Flagship model: Software engineering, code review, agentic coding	400K input / 128K output	25% faster, best agentic coding, $1.75/$14.00
`gpt-5.3-codex-spark`	Research preview: ultra-fast inference via Cerebras	400K input / 128K output	1000+ tokens/s, experimental
`gpt-5.2-codex`	Code review, security analysis	400K input / 128K output	79% SWE-bench Pro
`gpt-5.1-codex` ⭐	Software engineering, agentic coding workflows	400K input / 128K output	76.3% SWE-bench, $1.25/$10.00
`gpt-5.1-codex-mini`	Cost-efficient coding (4x more usage allowance)	400K input / 128K output	Near SOTA performance, $0.25/$2.00
`gpt-5.1-thinking`	Ultra-complex reasoning, deep problem analysis	400K input / 128K output	Adaptive thinking depth, runs 2x slower on hardest tasks

References

Load CLI reference and code review patterns:

Glob(pattern: "**/sdlc/**/skills/codex/references/codex-cli-reference.md", path: "~/.claude/plugins") → Read result

Arguments

$ARGUMENTS