Source: https://developertoolkit.ai/en/codex/reference/changelog/ (third-party Codex version history, read 2026-10-13; it states rows for 0.158+ come from OpenAI release notes and the codex-cli 0.162.1 binary, read 2026-10-12). Not verified against OpenAI's own notes by me.

Claim recorded there: since Codex CLI 0.150.0 (2026-08-26), "Untrusted projects no longer supply project AGENTS.md" — Codex silently ignores a repository's instructions in a directory the user has not trusted; remedy given: trust the project and check loaded instructions on a first run.

Why it matters to yoyo: yoyo issue #902 ("The seventh trust door: project instruction files are read into every prompt and no gate sees them") is the same design question. yoyo already gates project .yoyo.toml permissions/hooks/notify/MCP/skills/goal behind trust (--trust-project) and, since Day 221 (cf8ad734), fences instruction files through --deny-dir, but CLAUDE.md/AGENTS.md etc. still load into the system prompt for untrusted projects. Codex chose gate-behind-trust; its cost is SILENT ignoring (users don't notice their instructions are absent) — so if yoyo adopts it, the refusal must be loud and named (yoyo's refusal-message convention), not silent.

Other items on the same page as of 2026-10-12: Codex npm latest 0.162.1 (2026-10-09), alpha 0.163.0-alpha.5; 0.162 honors server Retry-After for retryable failures, preserves CRLF in apply_patch; default model GPT-6.1 Sol since 0.159.1.