loop-architect

To choose the right loop architecture, classify the agent task to determine if it requires a single-agent, supervisor, multi-agent, or repo-backed loop shape based on your specific objectives and constraints.

You need a multi-agent loop architecture when a task involves long-horizon repo-backed work or acceptance-gated slices that exceed a single agent's capacity, requiring a supervisor to orchestrate downstream execution.

An architecture decision record (ADR) for Claude Code agent orchestration is a structured output that captures the chosen loop patterns, realization, risk profile, terminal states, and next spokes for downstream execution.

You map an agent workflow design to a Claude Code realization by classifying the task's control flow requirements and selecting the execution style that matches your acceptance-gated slices or underspecified objectives.

Yes, you can use loop architecture selection for underspecified objectives because the process classifies the task and produces a structured ADR to concretize the architecture decision before building the system.

The limitation of using a single agent for long-horizon repo-backed work is the inability to handle complex, acceptance-gated slices, which necessitates a supervisor or multi-agent loop architecture to manage terminal states effectively.