# 0.1.8: Review recovery controls and next steps

Buttons and commands share recovery checks for the current Task, requirements/plan versions, node attempt, failed job and deadline. An eligible original job offers Retry review; any expired fault, explicitly non-retryable fault or stale artifacts offer Start a new review. A new job preserves the old failure and receives a new Session and budget; it neither extends the old job nor grants task execution. Missing records or changed bindings offer no review recovery and explain that `/task edit <full objective>` starts planning again and clears approval.

For reviews interrupted by a user pause or Host restart, Resume task exposes the fault and Start a new review when the old deadline has elapsed. A durably submitted decision can still be applied after artifact validity checks, without another model request. Requirement changes, artifact checks and execution approval rules continue to apply.

When only saved records are loaded, the current task offers Connect Session. This uses the native Session service to execute `/task status`, sends no model prompt and starts neither a review nor task execution. Historical tasks remain read-only. Connection failures appear beside the button with a command fallback.

Review faults explain remedies for credentials, model configuration, provider requests, artifacts, snapshots/runners or internal failures. Instructions follow current controls and do not promise manual retry when recovery is prohibited. A user-decision pause asks for a decision in the primary conversation before resuming. Rejected recovery requests appear on the current review card and survive successful background refreshes; another control attempt clears the previous error.

This version does not change the durable record schema, automatically rerun failed reviews, relax evidence checks or publish to npm.

[中文](0.1.8.zh.md)