Claude Codex Duo: Evidence-Based Review, Debate, and Planning

Описание и применение

Рабочий процесс Claude Codex Duo обеспечивает структурированный, доказательный обзор кода, дебаты и планирование с использованием OpenAI Codex, включая проверку доказательств и финальные рекомендации по слиянию.

Задача

Developers need structured, evidence‑based code review, technical debate, and planning workflows that avoid blind agreement between models and enforce verification of findings.

Результат

The workflow enables structured, evidence‑based code reviews, debates, and planning with clear audit trails, verification steps, and final merge recommendations.

Рабочий процесс

• 1. Install plugins via Claude marketplace
• 2. Run /codex-pr-review on PR or local changes
• 3. Lead and Codex agents produce concurrent reviews
• 4. Join gate seals artifacts and creates reconciliation ledger
• 5. Verify P0–P3 findings with evidence (reproduction, trace, tests)
• 6. Generate final report with merge recommendation
• 7. For debate, run /codex-debate with chosen mode
• 8. Record claims, evidence grades, and final ruling
• 9. For planning, run /codex-deep-plan with request
• 10. Produce evidence‑backed plan or answer

Доказательства

The project’s defining rule is that agreement between models does not establish whether a claim is correct.

Источник

Each claim receives an ID, owner, evidence grade, and status.

Источник

The workflow includes citation checks that resolve source references at the pinned commit and compare quoted text.

Источник