What problem does it solve? Multi-agent AI kits need model assignments per role, but hardcoding model names into skills creates vendor lock-in and silently breaks when hosts, plans, or allowlists change. This Skill discovers which models are actually reachable, ranks them into capability tiers with recorded evidence, and produces a verified role-to-model mapping. ## Core Features & Use Cases - Guided discovery and ranking: Enumerates reachable models per host (Kilo, Claude Code, others), ranks them into tiers using recorded evidence (host metadata, provider tier naming, model self-knowledge, or operator input), and proposes a four-profile mapping with justifications. - Persistent configuration: Writes the confirmed mapping to ~/.config/ai-architect-executor/roles.toml with host, model, effort, basis, and last_verified fields, without ever writing another tool's own config files. - Deterministic validation: Ships scripts/validate_roles.py to check that every skill declares a valid profile, no skill declares a model, all four roles are populated, and effort/model combinations are valid. - Use Case: After installing the Architect/Executor kit on a new machine, run the guided flow to detect available models, confirm a mapping of deep-reasoning to a high tier and bulk-execution to a cheap fast tier, then re-verify 24 hours later when the organization allowlist changes. ## Quick Start Ask the AI to run the interactive role setup to discover available models and configure the four role profiles in roles.toml.
