Release v0.16.0 · zenml-io/kitaru-skills

What changed

kitaru-adapter-builder now offers the shipped importer-backed adapter when a Python entrypoint already reports to Langfuse, Braintrust, LangSmith, Logfire, or Arize Phoenix, before proposing a custom in-process adapter.
kitaru-investigation, kitaru-importer-builder, and kitaru-replay-experiment distinguish this provider-trace route from static imports and custom adapters.
Replay guidance reads an agent version's runtime_capabilities before proposing a configuration. It blocks model/tool overrides and intercepting tool policies that an importer-backed adapter cannot apply, and states the remaining live-passthrough risk.

kitaru-replay-experiment uses baseline_evaluation_mode (--baseline-evaluation-mode) and documents the none / if_missing / force trade-offs, including parameter-aware if_missing reuse.
Result interpretation covers replay-linked, evaluator-version-specific aggregates.
kitaru-investigation and onboarding guidance document evaluator parameter provenance and float score-scale fields, aligned with zenml-io/kitaru#951.

kitaru-hosted-onboarding-tour no longer assembles dashboard URLs. It supplies one of five semantic navigation targets, exact resource IDs, and a short handoff message; the frontend builds the workspace-aware route (see zenml-io/zenml-frontend-monorepo#485).
Removed fallback route examples that could encourage URL invention.

Guidance-only release; no new skills or renamed commands. Reinstall or update the kitaru plugin to pick up the new routing and replay safeguards.

Full changelog: v0.15.1...v0.16.0