OpenAI's Codex ships free Auto-Review: a guardian agent, no quota hit
Day 2 of a 28-day Codex pledge to ship a feature or grant free usage every day. Also today: OpenAI's API tiers collapse from five to three.
Copy markdown
A second agent now double-checks the first
Auto-Review runs a guardian agent that watches your primary Codex agent, halting high-risk actions and keeping work on-spec—so a long agent run no longer means approving every step. It's free and doesn't count against plan usage; turn it on under Settings → Permissions → Auto-Review.
The 28-day 'improve or reset' pledge
Codex lead Thibault Sottiaux pledged on Oct 4 that for 28 straight days the team ships one clear improvement for most Codex/Work users each day—or, on a miss, hands out a full usage reset (free quota). Day 1 was a ~50% speedup; Auto-Review is Day 2. The run lands around end of October.
API usage tiers collapse from five to three
OpenAI consolidated its API tiers into Build, Launch, and Grow; orgs now auto-advance as cumulative credit purchases grow, and the change reshuffles rate-limit visibility per model. If you've been hitting a rate ceiling, check which tier you landed in.
Codex 0.160.1 unbreaks Windows MCP servers
The patch preserves SYSTEMROOT, TEMP, and TMP when launching remote stdio MCP servers, fixing a cross-platform breakage that bit Windows users running MCP tools. Update if your MCP setup has been flaky on Windows.