Claude skill to run a codex review

info.yaml

name: codex-prepush-review
description: >
  Run Codex to perform a pre-push code review for a given GitHub issue.
arguments:
  issue_number:
    type: integer
    description: GitHub issue number to review
tools:
  - bash
entrypoint: run.sh

run.sh

#!/usr/bin/env bash
# codex-review.sh
#
# Usage:
#   ./codex-review.sh 2
#
# Runs: codex exec "<prompt that tells codex to read issue #N + run git status/diff>"
#
set -euo pipefail

if [[ $# -ne 1 ]]; then
  echo "Usage: $0 <issue-number>" >&2
  exit 1
fi

ISSUE_NUM="$1"
if ! [[ "$ISSUE_NUM" =~ ^[0-9]+$ ]]; then
  echo "Error: issue-number must be an integer (got: $ISSUE_NUM)" >&2
  exit 1
fi

PROMPT=$(
  cat <<EOF
Look at issue #$ISSUE_NUM and then do a git status and where necessary git diff.

Goal: find likely bugs, missing edge cases, type-safety issues, Next.js server/client boundary mistakes, security issues, and missing tests.
Be strict on correctness; avoid large refactors unless necessary.

Output format:
1) Blockers (must-fix before push)
2) Important (should-fix)
3) Nits (optional)
4) Missing tests (specific test cases)
5) Questions for the author (only if truly needed)
EOF
)

# Run codex in non-interactive mode, save raw output, and print filtered results.
RAW_OUT="$(mktemp -t codex-prepush-review.XXXXXX)"
/home/dror/.npm-global/bin/codex exec --json "$PROMPT" | awk -v f="$RAW_OUT" '{
  print > f
  c++
  printf "\rlines: %d", c > "/dev/stderr"
} END { print "" > "/dev/stderr" }'
cat "$RAW_OUT" | jq -r 'select(.type=="item.completed" and .item.type=="agent_message") | .item.text'

SKILL.md

Codex Pre-Push Review

Use this skill when instructed by the user, typically when finished with implementing an issue.

Purpose

Run OpenAI Codex locally to review changes related to a GitHub issue. Codex will inspect the issue, run git status / git diff as needed, and return a structured review.

When to use

When instructed by the user

How it works

• Takes a GitHub issue number as input
• Executes a local Codex CLI command
• Writes raw JSON output to a temporary file
• Prints only the filtered review text (agent message output)

Expected output

1. Blockers (must-fix before push)
2. Important (should-fix)
3. Nits (optional)
4. Missing tests
5. Questions for the author (if needed)

Do not modify the script behavior. Address blockers before continuing.

https://gist.github.com/drorm/7851e6ee84a263c8bad743b037fb7abc