Why Your Recurring Agent Task Dies Right Before Delivery

If your recurring agent task keeps dying seconds before it posts results, the bootstrap is eating your timeout. A richly configured agent burns 60 to 120 seconds on memory loading, credentials, and skill discovery before it ever makes its first tool call. Everything after that competes for what's left of the budget, and the last step, usually the Slack post or email your team actually sees, is the one that gets clipped. The fix Size your timeout as bootstrap_p95 + work_p95 + buffer , not just th