What problem does it solve?
Choosing which AI model should execute each task is usually guesswork or hardcoded configuration. This Skill provides a default routing table that maps task complexity lanes (low, medium, high, critical) to the cheapest capable executor, derived from the models actually installed on the machine rather than copied examples.
Core Features & Use Cases
- Complexity lane semantics: Defines four routing lanes (low, medium, high, critical) matching the complexity vocabulary written into task frontmatter by cy-create-tasks.
- Catalog-driven derivation: Uses compozy__provider_models_list with cost fields (input_per_million / output_per_million) as the only source of concrete provider and model IDs, so routing never targets uninstalled providers.
- Stored workspace overrides: Writes runtime_rules via compozy__loop_configure as a per-workspace override for implement-tasks, with id > type > complexity matching precedence, and supports surgical id rules for escalation and reclassification.
- Use Case: When bootstrapping a batuta workspace, derive a routing table from the local provider catalog, present it with costs for operator confirmation, then store it so every run-loop child resolves the right model automatically.
Quick Start
Ask the agent to derive a cost-based routing table from the live provider catalog and store it as the workspace loop configuration for implement-tasks.
