cross-llm-pair-review

Cross-LLM code review improves compliance by coordinating Codex and Claude models to evaluate changes before commits, enforcing ADR governance policies and preventing unreviewed or irreversible edits to protected paths.

Set up cross-LLM review by running the inject-agent-context.sh script with pair-mode flags, designating Codex as the coder and Claude as the reviewer to evaluate edits before application.

Audit logging provides a tamper-evident record of all agent spawns, edits, and ceremonies, ensuring full traceability for compliance and governance in software development workflows.

A review rail protects code paths by requiring a second LLM, such as Codex, to review and approve edits made by Claude to protected paths before any changes are applied, preventing irreversible modifications.

Yes, this process requires specific ADRs (adr-052, adr-105, adr-106, adr-107, adr-108, adr-111) to enforce governance policies, validate rubric violations, and coordinate the cross-LLM review workflow effectively.

The limitation of using multiple LLMs for code review is the strict dependency on coordinating both Codex and Claude models, alongside specific ADR configurations, which restricts use in environments lacking these exact dependencies.