v0.7.2

Changed

• Codex CLI provider now defaults to gpt-5.4 so ChatGPT-authenticated local Codex installs work without extra model config
• bounded runtime gives Codex longer per-phase request caps for planner-style council runs that exceed the generic 15s draft timeout
• Codex quickstart docs now note the compatibility default and when to opt into *-codex variants explicitly

Fixed

• Codex CLI provider now rewrites incompatible *-codex model names when local login status shows Logged in using ChatGPT
• Codex CLI error reporting now preserves the meaningful trailing ERROR: lines instead of truncating the banner and hiding the real cause
• provider error classification no longer treats generic invalid_request_error responses as auth failures by default