cdx

Delegate scoped repository tasks to the OpenAI Codex CLI as a supervised headless subagent.

Updated May 10, 2026
One-click install
npx skills add https://github.com/cookkie03/skills --skill cdx-cookkie03
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: cdx
Source: https://github.com/cookkie03/skills/tree/main/cdx
Command: npx skills add https://github.com/cookkie03/skills --skill cdx-cookkie03

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running the Codex CLI headlessly requires correct flag usage, sandbox scoping, and output verification; this Skill provides the briefing templates, command patterns, and failure checks to delegate bounded coding work to Codex while you orchestrate and verify results. ## Core Features & Use Cases - Headless Task Delegation: Run codex exec with self-contained briefs, sandbox modes, and captured logs for implementation, refactors, and test triage. - Review-Only Passes: Use codex review with --uncommitted, --base, or --commit to get second opinions on explicit diffs. - Structured Output: Capture final messages with -o, stream events with --json, and validate results with --output-schema. - Use Case: Ask Codex to fix a failing test in a specific module while you supervise; the Skill ensures the brief is scoped, logs are captured, and the resulting diff is verified against the requested scope before reporting. ## Quick Start Use the cdx skill to delegate a bounded code change in my repository to the Codex CLI and report the verified diff and test results.

Frequently Asked Questions about cdx

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run Codex CLI headlessly for a coding task?

Use `codex exec -C /path/to/repo --sandbox workspace-write -o /tmp/final.txt` with a self-contained brief describing the goal, scope, and verification steps. Capture stdout and stderr separately, then verify the exit code, final message, and git diff before trusting the result.

What is the difference between codex exec and codex review?

codex exec performs non-interactive task execution such as edits, analysis, and test triage, while codex review is review-only and needs an explicit diff target via --uncommitted, --base <branch>, or --commit <sha>.

Which sandbox mode should I use with codex exec?

Use `--sandbox read-only` for analysis tasks and `--sandbox workspace-write` for bounded edits. Ask the user before using danger-full-access or bypass flags, and add extra writable roots with --add-dir only when genuinely needed.

Why does codex exec exit immediately with an empty output file?

Immediate exits with empty output usually mean bad flags prevented startup, so inspect stderr and logs first. Top-level codex flags and codex exec flags differ; only pass flags listed by `codex exec -h`, and never add --ask-for-approval to codex exec.

Can I run multiple Codex CLI jobs in parallel?

Yes, but only for independent chunks of work. Run parallel Codex jobs on non-overlapping scopes and reconcile any overlapping edits manually afterward.

What should I do when Codex CLI authentication fails?

Stop the run and ask the user to complete authentication interactively by running `codex login`. Headless runs cannot complete the login flow on their own.