codex-review

name	codex-review
description	Request an automated code review from OpenAI Codex CLI. Runs codex review against a base branch and summarizes findings.
argument-hint	[--base <branch>]
user-invocable	true

Codex Review

Request an external code review via the Codex CLI (OpenAI) and summarize findings.

Invocation

Input	Action
/codex-review	Review current branch against origin/development
/codex-review --base origin/main	Review current branch against specified base

Prerequisites

Verify Codex CLI is installed before proceeding:

which codex

If not found, ABORT with:

ERROR: Codex CLI not found.

Install: npm install -g @openai/codex
Docs: https://github.com/openai/codex

Aborting.

Workflow

Step 1: Gather Context

CURRENT_BRANCH=$(git rev-parse --abbrev-ref HEAD)
BASE_BRANCH="${USER_BASE:-origin/development}"

Verify there are changes to review:

git diff "$BASE_BRANCH"...HEAD --stat

If no changes, STOP with: "No changes between $CURRENT_BRANCH and $BASE_BRANCH. Nothing to review."

Step 2: Build Review Title

Extract Linear issue ID from branch name if present (pattern: INT-\d+).

Title format: "[INT-XXX] branch-description" or just "branch-description"

Step 3: Run Codex Review

CRITICAL: --base and [PROMPT] cannot coexist. Model and reasoning are set via -c config flags, not direct flags.

codex review \
  --base "$BASE_BRANCH" \
  -c model=gpt-5.4 \
  -c reasoning.effort=xhigh \
  --title "$TITLE" \
  2>&1 | tee /tmp/codex-review.txt

Step 4: Read and Summarize

Read /tmp/codex-review.txt and present a summary to the user:

1. Overall verdict — what did Codex think of the changes?
2. Key findings — list actionable items (bugs, style issues, suggestions)
3. File-by-file highlights — only if Codex provided per-file feedback

Do NOT editorialize or add opinions beyond what Codex reported. Present the findings factually.

Edge Cases

Situation	Action
Codex not installed	Abort with install instructions (see Prerequisites)
No changes to review	Stop with informational message
Codex command fails	Show full error output, suggest checking OPENAI_API_KEY
Very large output (>500 lines)	Summarize top findings, note full output at /tmp/codex-review.txt

Critical Rules

1. NEVER modify code based on Codex findings unless the user explicitly asks
2. NEVER pass a prompt argument alongside --base — they conflict
3. ALWAYS save raw output to /tmp/codex-review.txt for reference
4. ALWAYS use -c flags for model and reasoning config, not direct flags