v1.4.3

Changes

• Adds a one-shot soft retry for ChatGPT Codex cyber_policy false positives on Fast/Priority tier by retrying the same payload once without service_tier=priority.
• Keeps policy behavior intact: no prompt rewriting and no policy bypass.
• Updates GPT-5.5 model limits to keep a 400k input compaction budget with 130k output headroom.

Install

opencode plugin @guard22/opencode-multi-auth-codex@latest -g -f

Verification

• npm run build
• npm test
• npm pack
• clean npm install smoke test from registry @latest