dsh-agent-dispatch

When a single main agent handles every task, analysis, review, and debugging all pile into one context window, a wrong model choice stalls the entire chain, and multi-step workflows become impossible. dsh-agent-dispatch lets one natural-language sentence route a task to a persona-defined agent sub-proxy with its own isolated context, keeping the main conversation clean and each sub-proxy focused.

Capabilities span three layers: routing assigns each agent a model priority table, automatically falling back to the next model on failure without manual retry; orchestration uses squad templates to decompose multi-angle goals into ordered steps, where dependent placeholders auto-inject upstream results and independent steps run in parallel; interaction supports three trigger styles, auto-routing in dialogue, a slash-command menu, and a draggable floating activity ball, while re-dispatching to the same agent reuses the existing sub-proxy for seamless follow-ups. All changes take effect immediately without restart.

Ideal for developers who split complex work across domain-specific agents, and for small teams that need auditable, pipeline-style multi-angle investigation across their projects.