error-recovery

Subagent failure handling uses retry strategies (simple retry, context enhancement, scope reduction) paired with escalation to specialized contractors. Track retry state and match the recovery strategy to the failure type—context issues trigger context enhancement, complexity triggers scope reduction, and unrecoverable failures escalate or abort with a report.

Error recovery implements five strategies: simple retry repeats the task, context enhancement adds detail to improve outcomes, scope reduction breaks the task into smaller steps, escalation routes to specialized contractors, and abort with report halts execution and documents the failure for review.

Escalate when retry limits are reached, the failure indicates a knowledge or capability gap, or the error pattern suggests the subagent cannot resolve it independently. Escalation routes the task to specialized contractors equipped to handle the specific failure class.

Retry state tracking counts attempts and evaluates failure patterns against configured limits. Once limits are met, the executor stops retrying and moves to escalation or abort, preventing loops and ensuring forward progress through the recovery strategy chain.

Yes. Error recovery integrates with the executor loop to apply strategies sequentially—start with simple retry, then context enhancement, then scope reduction. Each strategy builds on prior attempts; if all fail, escalate or abort based on failure severity and retry exhaustion.