The plan approval UI now shows context window usage against the execution model's actual capacity, preventing misleading displays when plan mode runs on a larger-context model than the execution model.
Ensures the context usage indicator in plan approval reflects the actual execution model's budget, disabling the 'keep context' option when projected usage exceeds 95%.
Developers using the coding agent get an accurate preview of context window pressure before approving a plan, so they can avoid starting a plan that will immediately hit context limits during execution. This prevents confusing '0.0%' readings after aborted approvals and stops users from approving plans that would fail due to insufficient context in the execution model. Watch for whether the 95% threshold is optimal for various workflows and if edge cases in context baseline calculation are fully resolved.