johngorat
UserClaude Code skills that wire OpenAI Codex in as an adversarial code reviewer — task, debate, loop until APPROVED
Categories
Indexed Skills (3)
codex-check
One-shot Codex review of the current diff — a single round, no debate loop. Claude triages the findings, fixes the real ones, and recommends escalating to /codex-debate if a major finding stands. Use for routine changes on proven patterns, small diffs, config/docs edits, or a quick second opinion where a full debate is too heavy.
codex-debate
Run a task, then have Codex (top available GPT model, e.g. GPT-5.6 Sol) adversarially review the resulting diff, debate the findings, fix what is real, and loop until APPROVED. Use when the user asks for a task with codex review/debate, a second-opinion review loop, or independent verification of a change.
codex-plan
Turn a substantial multi-stage task into a staged, gate-reviewed execution plan — stages with deliverables, a review tier per stage (codex-check vs codex-debate), past reviewer findings converted into hard requirements, and named tools to reuse. Use BEFORE writing any code when starting a port, migration, new pipeline run, or any task big enough to need more than one review.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.