Question
How should the primary controller route work across strong and weak models when the weak model needs more explicit skill and tool guidance?
Resolve capability matching, required versus optional skills, model-tier selection, fallback and substitution, default recursion depth, deeper research permissions, cache admission, and the evidence required before a worker result can be promoted. Optimize for useful work and cache reuse without allowing an unbounded Agent tree or silent capability downgrade.
Question
How should the primary controller route work across strong and weak models when the weak model needs more explicit skill and tool guidance?
Resolve capability matching, required versus optional skills, model-tier selection, fallback and substitution, default recursion depth, deeper research permissions, cache admission, and the evidence required before a worker result can be promoted. Optimize for useful work and cache reuse without allowing an unbounded Agent tree or silent capability downgrade.