mirror of
https://github.com/block/buzz.git
synced 2026-11-02 07:22:55 +00:00
getProviderEffortConfig() is now generated-backed (thin wrapper over resolveModelCapabilities() from modelCapabilities.ts). The old hand-table implementation is preserved as getProviderEffortConfig_oldHandTable() for the differential harness shim only — nothing user-facing imports the _old name. AgentConfigFields.tsx and buzzAgentModelTuningFields.tsx continue calling getProviderEffortConfig() with zero call-site churn and now receive generated capabilities (including the F1 correction: databricks-gpt-5-5 allows only low|medium|high). resolveModelLabel() gains an optional provider param that routes through provider-qualified resolveModelCapabilities() registryLabel before falling back to the unscoped DATABRICKS_MODEL_NAMES map. Provider threaded at all provider-bearing call sites: ManagedAgentRow, ModelPicker, TeamIdentityCard, usePersonaModelDiscovery, agentCardModelLabel, UnifiedAgentsSection, and UserProfilePopover. AgentConfigFields fallbackModel label (genuinely providerless) is left unscoped per plan. Explicit .ts extension on modelCapabilities import in buzzAgentConfig.ts (allowImportingTsExtensions is on) — enables Duncan to delete the ts-esm-loader.mjs shebang bootstrap and run the differential script as plain node --experimental-strip-types. Test update: databricksModelNames.test.mjs regex updated to match the provider-threaded call site resolveModelLabel(model.id, model.name, agent.provider). All four gates at this head: typecheck clean, 3847/3847 unit pass, biome clean (2 pre-existing infos), check:file-sizes and check:pubkey-truncation pass. Co-authored-by: Will Pfleger <pfleger.will@gmail.com> Signed-off-by: Will Pfleger <pfleger.will@gmail.com>