autosearch:channel-selection

Group-first channel selection optimizes runtime routing by using a two-stage pick to select 1-3 groups and 3-8 leaf channels, preventing the AI from reading all 41 SKILL.md bodies. This reduces latency and computational overhead.

To implement two-stage channel selection, run the clarify step with a research query to prepare channel_priority hints, then execute the channel-selection step to pick groups and leaves before invoking run_channel.

Deterministic channel scoring in the v2 tool-supplier architecture requires a research query, clarify rubrics, and channel_priority hints. These inputs drive the selection of groups and leaves from the router index.

A two-stage group-first routing strategy replaces flat-rank selection across 41 channels to ensure scalable operation. It applies deterministic scoring, hard caps, and fallback rules to provide predictable latency and cost.

Yes, the channel selection mechanism includes fallback rules alongside deterministic scoring and hard caps. These rules ensure safe operation by maintaining scalable routing even when initial leaf channel choices are insufficient.