API 500 errors during subagent-heavy sessions

FIX YOUR API LOAD BALANCING

During a session with multiple sequential subagent dispatches (Task tool), the API repeatedly returned HTTP 500 errors. This happened 4+ times in a single session, killing subagent runs and requiring manual retries or fallback to direct implementation.

Date: 2026-02-25
Model: claude-opus-4-6
Pattern: Back-to-back Task tool invocations (general-purpose and superpowers:code-reviewer subagents). Failures clustered — once one 500 hit, subsequent calls would also 500 for a period.

Impact: Significant workflow disruption. Subagent-driven development relies on dispatching implementer + reviewer agents per task. When the API drops requests, the entire workflow stalls.

Request: Improve load balancing and retry behavior for sustained multi-agent sessions.