v2026.7.28-2

New Features

• Self-correcting goal continuations — Detect goals that repeatedly wake only to wait on the same active monitors, then require an explicit process-health and blocked-state audit before another wait cycle (#443).
• Clearer live monitor status — Show monitor activity with a forward count and visible glyph so users can recognize active background subscriptions at a glance (#441).
• Leaner Grok 4.5 execution prompt — Reduce redundant CEO-core instructions while preserving the intent gate, autonomous delivery contract, verification discipline, and stop conditions (#442).

Breaking Changes

Added

• Add a goal-continuation stall check after three consecutive monitor-driven wakeups, including reset behavior for user input, monitor completion, goal replacement, and session lifecycle changes (#443).

Changed

• Render the interactive monitor footer with a count-forward label and activity glyph for faster recognition in narrow and busy terminal layouts (#441).
• Diet the Grok 4.5 CEO prompt preset by removing duplicated instructions without weakening its outcome-first execution, manual-QA, failure-recovery, and binding stop contracts (#442).

Fixed

• Preserve Anthropic conversations with compacted or otherwise missing tool-use blocks by demoting orphaned tool-result references before request submission (#437).
• Treat blank, whitespace-only, and null update_goal reasons as omitted when completing a goal, while continuing to reject non-empty completion reasons and requiring a meaningful blocked reason (#440).
• Execute a new eval cell when a terminal cell ID is reused and suppress unavailable timing metadata, preventing stale result replay and misleading duration output (#439).

Removed