sadd:launch-sub-agent

Sub-agent orchestration dispatches focused agents with isolated context to enable targeted task execution and clean cross-task boundaries. This pattern uses model selection and prompt orchestration to handle architecture, research, and development tasks deterministically.

Zero-shot CoT reasoning is a prompt orchestration mechanism used to analyze tasks and determine optimal model configurations. It constructs sub-agent prompts with a zero-shot CoT prefix and a mandatory self-critique suffix to verify execution.

To launch a sub-agent, select a model, add optional specialized agent prompts, and dispatch the sub-agent using the Task tool. The prompt must include a zero-shot CoT prefix for reasoning and a mandatory self-critique suffix for verification.

Yes, multi-agent orchestration is applicable to architecture, research, and development tasks across codebases and documentation work. It isolates context for each phase, ensuring clean boundaries during cross-task execution.

No, specialized agents are optional. The core requirement is phase-based orchestration using zero-shot CoT reasoning to determine the optimal model configuration and construct the sub-agent prompt for the given task.

Context isolation ensures targeted task execution and clean cross-task boundaries. By dispatching focused sub-agents through the Task tool, the orchestration pattern prevents context bleed and maintains deterministic execution.