Skip to main content
Run a Codex review that questions whether your chosen approach is the right one — not just whether it is implemented correctly. The review challenges assumptions, surfaces tradeoffs, examines failure modes, and asks whether a different approach would have been safer or simpler. This command uses the same review target selection as /codex:review, including --base <ref> for branch review. Unlike /codex:review, it accepts extra focus text after the flags to steer the review toward specific areas of concern.

Syntax

Flags

boolean
Run the review in the foreground and wait for results.
boolean
Run the review as a background job. Check progress with /codex:status.
string
Compare against this git ref. Accepts a branch name, tag, or commit hash — for example, main.
string
Controls what is included in the review. Accepted values:
  • auto (default) — reviews the working tree if dirty, otherwise compares against the detected default branch
  • working-tree — always reviews uncommitted changes
  • branch — compares against the auto-detected default branch (main, master, or trunk)
string
Optional positional text added after the flags. Use this to give the review a specific focus — for example, challenge whether this was the right caching and retry design.

Examples

When to use it

  • Before shipping, when you want to challenge the direction — not just catch implementation bugs
  • When you want focused scrutiny on design choices, tradeoffs, and hidden assumptions
  • When you want to pressure-test specific risk areas like authentication, data loss, rollback behavior, race conditions, or reliability under failure

Difference from /codex:review

/codex:review maps directly to the built-in Codex reviewer and does not support custom focus text. /codex:adversarial-review accepts extra focus text and frames the review around whether the approach itself is sound, rather than whether the implementation is correct.
This command is read-only. It does not fix code or apply any changes to your repository.