[Feature Request] Long-running tasks need persistent state and decision guardrails to prevent session abandonment and course reversal

Bug Description
I understand without full graph or MCP driven approach, the decisions and path will waver. The session should not be able to commit suicide and pivot from a 30+ hour endeavour ( were making progress on ) because it screwed something up and then signals sending alarming information. Never ever should it swerve 180 degrees, let alone 90. If it thinks has to, should ask the user.. as usually does early in a session. This has been compacted twice but shit, IF that does not actually work yet, we should have a graph or other permanent store for VERY long running and complex tasks so that a fresh session can pickup and have full health. Telling the user to do that instead Tyvm >:(.

Environment Info

• Platform: darwin
• Terminal: xterm-256color
• Version: 2.1.220
• Feedback ID: 8c73d104-4c90-4a60-aee0-f164d5b19735

Errors

[]