Codex — независимое ревью и отладка кода

★ 7.0 · vibe-coding

Codebase and Architectural Context is a Claude Code skill that drives the codex CLI agent (GPT-5.6-sol) as a delegate for independent code analysis: adversarial review of plans and designs, pre-PR code review, root-cause debugging, test writing, and well-scoped implementation tasks. Reach for it when you need a fresh pair of eyes before opening a PR, when a bug reproduces but the cause is unclear, when you want your architecture doc attacked with concrete counterexamples, or when tests need to be added without interrupting the main session. The skill covers `codex exec` invocation with `-m gpt-5.6-sol`, `-c model_reasoning_effort=xhigh`, `--dangerously-bypass-approvals-and-sandbox`, and a mandatory `-o` flag to capture output; multi-round session patterns via `resume` and `fork`; ready-made recipes for review, debug, test writing, and implementation; and git worktree isolation rules to prevent concurrent runs from clobbering each other's workspace. The direction is outbound delegation to an external agent, not a replacement for the main Claude Code session — the independence is the point: codex had no part in writing what it reviews, so it catches the current session's blind spots. Codex output is a hypothesis, not a verdict; every concrete claim requires an empirical probe before acting on it.