Workflow Retry Design with Claude Code: Human Review Queue
A production playbook for workflow retry design in cross-industry operations using Claude Code: human review queue, run-scoped inputs, logs, typed results, and artifacts.
Audience: Backend engineers building reliable agents
The problem
Backend engineers building reliable agents need workflow retry design to run repeatedly against terminal states, retry policies, idempotency keys, and failure logs. In cross-industry operations, the pain is not one good answer; it is repeatability, auditability, exception handling, and evidence that survives handoff.
Implementation path
Split the workflow retry design result into automatable fields and review-only exceptions, then send low-confidence cases to a human queue with evidence artifacts attached.
Tradeoffs and failure modes
Human review slows a subset of runs, but it lets the workflow ship before every edge case is fully automated. For workflow retry design, the practical test is whether a second run can be debugged, retried, and consumed by a product without reading the raw agent transcript.
Review handoff
review_status: needs_review | approved | rejected
review_reason: string
source_evidence: artifact_url[]
agent: Claude Code
workflow: workflow-retry-design
Run this on Argo