subagent-verify-applytest-explore

Subagent orchestration automates a four-phase cycle—verify, triage, apply-and-test, and explore—until verification passes clean. It coordinates automated handoffs across GLM, Kimi, and Gemma components, enforcing a maximum of three full iterations to resolve code failures.

Git commits during agent handoffs are enforced at every phase transition before proceeding to the next step. This requires the workflow to commit changes to Git after verify, triage, apply-and-test, and explore phases, maintaining a fully traceable history throughout the iteration cycle.

Tmux-based orchestration coordinates multiple AI agents by managing defined agent models across GLM, Kimi, and Gemma components. It facilitates looping diagnostics and automated handoffs within a terminal environment, driving the workflow toward a clean verify result.

The maximum number of iterations allowed for automated code verification workflows is three full cycles. This iteration limit constrains the verify, triage, apply-and-test, and explore phases to prevent infinite looping while attempting to reach a clean verify outcome.

Automated code verification fails to pass clean when the explore phase identifies unresolved failures after applying fixes. The workflow handles this by looping back through triage and apply-and-test phases, but it halts if the maximum of three full iterations is reached without success.

Subagent workflow automation supports automated triage for code issues by passing verification failures to a dedicated triage phase. This phase analyzes the issues before handing off to the apply-and-test phase, ensuring failures are systematically diagnosed and addressed within the iteration limit.