Claude Code CLI Codex Skill

Updated as of June 25, 2026.

A Codex skill for delegating work to the local Claude Code CLI without burning tokens on flag guessing. It gives future Codex agents direct guidance for claude -p, JSON output, --max-budget-usd, --max-turns, model aliases, reasoning effort, ultrathink, ultracode, resumable sessions, background Claude agents, tool restrictions, MCP restrictions, and workflow orchestration.

Search terms this repo is meant to satisfy: Claude Code CLI Codex skill, Claude skill for Codex, call Claude from Codex, Claude Code orchestration skill, Claude CLI --resume skill, Claude ultracode skill, Claude ultrathink skill, Claude reasoning effort skill, Claude Code JSON print mode skill.

What Is Included

• claude-code-cli/SKILL.md: the installable Codex skill.
• claude-code-cli/references/cli-options.md: current option notes with official source links.
• claude-code-cli/scripts/claude_call.py: a small wrapper for low-waste print-mode calls.
• claude-code-cli/agents/openai.yaml: UI metadata for Codex skill surfaces.

Install

Clone this repository, then copy or symlink the claude-code-cli directory into your Codex skills directory:

git clone https://github.com/Michaelihc/claude-code-cli-codex-skill.git
Copy-Item -Recurse .\claude-code-cli-codex-skill\claude-code-cli "$env:USERPROFILE\.codex\skills\claude-code-cli"

Then invoke it in Codex with:

Use $claude-code-cli to ask Claude Code for a second opinion on this design.

Quick Example

python .\claude-code-cli\scripts\claude_call.py --prompt "Reply with exactly: ok" --budget 0.05 --result-only

For a resumable Claude thread, keep the full JSON output so you can read session_id:

$first = claude -p --output-format json --model haiku --effort low --max-turns 1 --max-budget-usd 0.05 --tools "" --disallowedTools "mcp__*" --strict-mcp-config "Remember alpha and reply ready."
$sid = ($first | ConvertFrom-Json).session_id
claude -p --resume $sid --output-format json --model haiku --effort low --max-turns 1 --max-budget-usd 0.05 --tools "" --disallowedTools "mcp__*" --strict-mcp-config "What token did I ask you to remember?"

Accuracy

This skill was written against Claude Code CLI 2.1.191 and official Claude Code docs available on June 25, 2026:

• https://code.claude.com/docs/en/cli-reference
• https://code.claude.com/docs/en/sessions
• https://code.claude.com/docs/en/model-config
• https://code.claude.com/docs/en/workflows
• https://code.claude.com/docs/en/skills

It intentionally distinguishes ultrathink from --effort, and ultracode from --effort. ultrathink is a prompt keyword for one-turn deeper reasoning; ultracode is a Claude Code workflow setting that uses xhigh plus dynamic workflow orchestration.

Validation

Validated locally on June 25, 2026:

• quick_validate.py passed for claude-code-cli/SKILL.md.
• python -m py_compile passed for scripts/claude_call.py.
• A live wrapper call returned exactly wrapper-ok.
• A separate Codex subagent used this skill to start a Claude Code print-mode session returning alpha, resumed it by session_id, and got beta from the same session. The tested CLI was Claude Code 2.1.191, authenticated via Claude Max; the model resolved to claude-haiku-4-5-20251001.

License

MIT.