Claude Skill MartBrowse skillsQuick linesLearn by videoTerminal guideWhat is a Skill?
 
← Back to list

Codex Panel A/B Experiment

Runs one holistic codex review against three focused passes (security, Ecto, LiveView) on the same branch diff, then classifies findings to prove whether the extra passes catch real misses.

Dev & CodingAdvanced★ 522⑂ 36AI score 8/10Last updated: Aug 11, 2026

What it does

  • Launches 1 holistic codex exec review plus 3 dimension-focused codex exec workers (security / ecto / liveview) in parallel on the same branch diff.
  • Classifies every focused finding as DUPLICATE, REAL MISS (genuine defect the holistic pass skipped), or FALSE POSITIVE.
  • Requires reading the actual code at file:line before a REAL MISS counts.
  • Emits a verdict table and persists it to .claude/reviews/codex-ab-{date}/VERDICT.md.
  • Enforces "Iron Laws" to protect your quota: fresh (never-reviewed) diffs only, exactly 4 codex runs, and never pipe the multi-thousand-line .log streams into context.

Who it's for

  • Elixir/Phoenix developers building or tuning a codex-CLI code review pipeline.
  • Teams that want to decide "does splitting review prompts actually help?" from measured data instead of intuition.
  • Tooling owners who need a recorded cost/benefit verdict (4× quota) for AI review strategies.

Examples

  1. On a brand-new feature branch with no prior review, run /codex-ab — about 5 minutes later you get 4 findings files and a classification table.
  2. Run /codex-ab develop to A/B a release branch against develop as the base.
  3. For a UI-heavy LiveView branch, add only the LiveView-focused pass (2× cost), re-test, and accumulate results in VERDICT.md before deciding whether to ship a --codex-panel mode.

· · · Install guide · · ·

Try it now, no install

Paste this into Claude to use the skill without installing anything.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/oliver-kriska/claude-elixir-phoenix/HEAD/.claude/skills/codex-ab/SKILL.md

What I want: (describe your task here)

If Claude can't open the link, open it yourself and paste the contents instead.

↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the .claude/skills/codex-ab folder from the GitHub repo oliver-kriska/claude-elixir-phoenix into my ~/.claude/skills/codex-ab/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/oliver-kriska/claude-elixir-phoenix.git && mkdir -p ~/.claude/skills && cp -r claude-elixir-phoenix/.claude/skills/codex-ab ~/.claude/skills/

⚠ This is a third-party skill. Check the source repository before installing.

  1. Clone the repo: git clone https://github.com/oliver-kriska/claude-elixir-phoenix.git
  2. Create the skills directory: mkdir -p ~/.claude/skills
  3. Copy the skill: cp -r claude-elixir-phoenix/.claude/skills/codex-ab ~/.claude/skills/
  4. Make the harness executable: chmod +x ~/.claude/skills/codex-ab/scripts/codex-panel-ab.sh
  5. Verify the codex CLI is available: command -v codex (install it first if missing).
  6. Start Claude Code in your Elixir/Phoenix project root and run /codex-ab or /codex-ab develop.
  7. Important: only run it on a branch that has never been codex-reviewed — a drained diff returns no findings and wastes quota.
View source on GitHub ↗License: MIT