Problem to solve
Catalog validate runs each skill serially. Large skill collections spend most of their wall-clock time in per-skill Tier 1 runs, and there is no --workers flag to scale throughput inside SkillEvaluator.
Proposed approach
Add --workers N to catalog mode (default 1 keeps today's serial behavior and per-skill pipeline view). Values above 1 validate skills in parallel child processes with isolated output_dir/<skill>/ trees and the same exit aggregation as serial mode.
Additional context
Validated gap on main. Orthogonal to #41 (Tier 2 vector index). Works with catalog-summary.json fleet rollup.
Before submitting
Problem to solve
Catalog
validateruns each skill serially. Large skill collections spend most of their wall-clock time in per-skill Tier 1 runs, and there is no--workersflag to scale throughput inside SkillEvaluator.Proposed approach
Add
--workers Nto catalog mode (default 1 keeps today's serial behavior and per-skill pipeline view). Values above 1 validate skills in parallel child processes with isolatedoutput_dir/<skill>/trees and the same exit aggregation as serial mode.Additional context
Validated gap on main. Orthogonal to #41 (Tier 2 vector index). Works with
catalog-summary.jsonfleet rollup.Before submitting