/codex:review vs /codex:adversarial-review: What's the Difference in OpenAI's Codex Plugin

Both commands run read-only Codex code reviews through the same plugin runtime, but /codex:adversarial-review explicitly challenges design decisions and trade-offs while /codex:review provides neutral defect detection.

The openai/codex-plugin-cc repository provides two slash commands for automated code review. Understanding when to use each helps developers get the right type of feedback—whether they need straightforward bug detection or aggressive scrutiny of architectural choices.

Core Purpose and Framing

The fundamental difference lies in how the model approaches your code.

/codex:review performs a neutral, standard inspection. It checks for implementation defects, style inconsistencies, and obvious bugs without预设 assumptions about your approach.

/codex:adversarial-review explicitly frames the review as a challenge. According to the prompt template in plugins/codex/prompts/adversarial-review.md, this mode encourages the model to surface design-level concerns, question trade-offs, and probe hidden assumptions that a neutral review might overlook.

Both commands share identical execution mechanics: they use the same diff-selection logic (working-tree, branch diff, or explicit --base), the same foreground/background execution modes, and the same prohibition against auto-fixing or patching code. See plugins/codex/commands/review.md and plugins/codex/commands/adversarial-review.md for the command definitions.

Command Syntax and Unique Features

Standard Review Command

The /codex:review command accepts these flags:

/codex:review --wait                    # Foreground execution

/codex:review --background              # Force background task

/codex:review --base <ref>              # Compare against specific branch/commit

/codex:review --scope auto|working-tree|branch   # Control diff selection

Critical limitation: This command does not accept free-form focus arguments. The review scope is determined entirely by flags and automatic diff detection.

Adversarial Review Command

The /codex:adversarial-review command supports the same flags plus optional trailing focus text:

/codex:adversarial-review --wait --base main
/codex:adversarial-review --scope branch --base develop performance security

In the second example, performance security becomes the focus argument passed to codex-companion.mjs. This directs the adversarial review to scrutinize those specific aspects, as documented in plugins/codex/commands/adversarial-review.md.

Runtime Execution: How Both Commands Work

Both commands invoke the same shared runtime script at plugins/codex/scripts/codex-companion.mjs. The difference is the subcommand passed:

Command	Companion Script Invocation
/codex:review	node …/codex-companion.mjs review "$ARGUMENTS"
/codex:adversarial-review	node …/codex-companion.mjs adversarial-review "$ARGUMENTS"

The execution flow is identical for both:

1. Size estimation: If no --wait or --background flag is provided, the plugin estimates diff size
2. User prompt: Presents "Wait for results" and "Run in background" options, marking one as (Recommended) based on estimated workload
3. Review execution: Launches the appropriate Codex review with command-specific prompt templates

Practical Usage Examples

Standard Review Workflows

Run a normal review immediately in the foreground:

/codex:review --wait

Let the plugin decide execution mode based on diff size:

/codex:review

Review changes against a specific base branch:

/codex:review --base main --scope branch

Adversarial Review Workflows

Challenge design decisions with immediate feedback:

/codex:adversarial-review --wait --base main

Focus the challenge on specific concerns:

/codex:adversarial-review --scope working-tree memory-usage thread-safety

Run adversarial review against branch diff with custom focus:

/codex:adversarial-review --scope branch --base develop api-consistency error-handling

Key Source Files

Path	Purpose
plugins/codex/commands/review.md	Standard review command definition and flag handling
plugins/codex/commands/adversarial-review.md	Adversarial review command with focus argument parsing
plugins/codex/scripts/codex-companion.mjs	Shared runtime entry point for both commands
plugins/codex/prompts/adversarial-review.md	Prompt template shaping the challenge-oriented tone
plugins/codex/prompts/stop-review-gate.md	Prompt template for neutral standard reviews

Summary

• Use /codex:review for straightforward defect detection, style checks, and bug identification
• Use /codex:adversarial-review when you want aggressive scrutiny of design decisions, trade-offs, and assumptions
• Add focus text to adversarial reviews to direct attention at specific concerns like performance, security, or API design
• Both commands share the same execution infrastructure in codex-companion.mjs and differ only in framing prompt and argument handling
• Neither command modifies code—both are strictly read-only review operations

Frequently Asked Questions

Can I add focus text to a standard /codex:review command?

No. The standard review command in plugins/codex/commands/review.md does not parse trailing free-form arguments. Only /codex:adversarial-review supports the optional focus … parameter to direct review attention. Use adversarial review if you need targeted examination of specific aspects.

Do both commands support background execution?

Yes. Both commands implement identical execution-mode logic. Without --wait or --background flags, the plugin estimates diff size and recommends either foreground or background execution, marking the suggested option as (Recommended). Use --wait to force immediate results or --background to always run as a Claude task.

What happens under the hood when I run these commands?

Both commands invoke plugins/codex/scripts/codex-companion.mjs with their respective subcommand (review or adversarial-review). The companion script handles Codex API interaction, while command-specific prompt templates—stop-review-gate.md for standard reviews and adversarial-review.md for adversarial ones—shape the model's output tone and framing.

When should I choose adversarial review over standard review?

Choose adversarial review when you need design-level critique rather than implementation-level bug detection. The adversarial framing in plugins/codex/prompts/adversarial-review.md explicitly instructs the model to challenge your approach, surface unstated assumptions, and evaluate trade-offs. Add focus arguments to concentrate this scrutiny on specific concerns like security boundaries, scalability, or maintainability.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}