name: oracle description: Run GPT-5.2 Codex (gpt-5.2-codex high) via OpenAI Codex CLI to review code for bugs, issues, and architecture problems. Use when the user wants a second opinion on code, requests a code review from an external model, says "ask the oracle", "get codex review", "run oracle", or wants GPT-5.2 to analyze their codebase for issues.

Oracle - GPT-5.2 Codex Code Review

Run OpenAI's Codex CLI with gpt-5.2-codex high to get an expert second opinion on code quality, bugs, and architecture.

Prerequisites

Codex CLI must be installed. If not available, install with:

npm install -g @openai/codex

Requires OPENAI_API_KEY environment variable to be set.

Usage

Use the codex review subcommand for non-interactive code reviews. The -m flag sets the model.

Review Specific Files

codex -m gpt-5.2-codex review "Review <file_path> for bugs, security issues, and architecture problems. Output format: file:line - [HIGH/MEDIUM/LOW] - description. Focus only on actual issues, not style."

Review Uncommitted Changes

codex -m gpt-5.2-codex review --uncommitted "Review these changes for bugs, security issues, and architecture problems."

Review Changes Against a Branch

codex -m gpt-5.2-codex review --base main "Review changes against main branch for bugs and issues."

Review a Specific Commit

codex -m gpt-5.2-codex review --commit <sha> "Review this commit for bugs and issues."

Prompting GPT-5.2 Codex Effectively

GPT-5.2 responds best to:

1. Explicit length constraints - Specify if you want brief or detailed analysis
2. Scope discipline - "Focus only on bugs and security, no refactoring suggestions"
3. Structured output requests - "List issues as: file:line - severity - description"
4. Grounded claims - Ask it to cite specific file paths and line numbers

Example Review Prompts

Quick bug scan:

Review this codebase for bugs and security vulnerabilities. Output format:
- file:line - [HIGH/MEDIUM/LOW] - description
Keep response under 500 words. Focus only on actual issues, not style.

Architecture review:

Analyze the architecture of this codebase. Identify:
1. Coupling issues between modules
2. Single points of failure
3. Scalability concerns
4. Missing abstractions
Be specific with file paths. No refactoring code, just identify issues.

Security audit:

Security audit this code. Check for:
- Injection vulnerabilities (SQL, command, XSS)
- Authentication/authorization issues
- Data exposure risks
- Dependency vulnerabilities
List each finding with file:line and severity rating.

Workflow

1. Determine what to review (specific files, uncommitted changes, branch diff, or commit)
2. Select appropriate review prompt from examples above or customize
3. Run codex -m gpt-5.2-codex review with the appropriate flags
4. Present findings to user with file paths and line numbers
5. Offer to help address any issues identified

CLI Reference

codex review [OPTIONS] [PROMPT]

Options:
  --uncommitted        Review staged, unstaged, and untracked changes
  --base <BRANCH>      Review changes against the given base branch
  --commit <SHA>       Review the changes introduced by a commit
  --title <TITLE>      Optional commit title to display in the review summary
  -m, --model <MODEL>  Model to use (e.g., gpt-5.2-codex)
  -c, --config <k=v>   Override config values (e.g., -c approval=never)

Reference

For detailed GPT-5.2 prompting patterns, see references/gpt52-prompting.md.