v0.1.4

[0.1.4] - 2026-04-20

Fixed

• expected_response and language_code generated by synthesizers are now serialized as direct fields on Prompt instead of being nested under metadata. Previously the backend silently dropped them during testSets().create(...), so generated tests had no expected response on the platform (#3).
• MultiTurnSynthesizer no longer throws NullPointerException when the LLM omits min_turns or max_turns from a generated test. The values are now treated as optional and set to null on TestConfiguration, matching the Python SDK's behaviour.
• Test.metadata now correctly deserializes from the backend's test_metadata field (renamed on the server to avoid colliding with SQLAlchemy's reserved Model.metadata). Before this change test.metadata() returned null after any testSets().getTests(...) round-trip, matching an analogous bug the Python SDK already worked around.

Changed

• Prompt record gained two top-level fields — expectedResponse (JSON expected_response) and languageCode (JSON language_code) — and is now @JsonInclude(NON_NULL) so unset fields no longer appear in serialized output.
• Removed the role field from Prompt. It was never part of the Rhesis platform API or the Python SDK and had no effect beyond bloating the JSON payload. The positional constructor now takes (id, content, expectedResponse, languageCode, metadata); callers using Prompt.builder() only need to drop any .role(...) call.

v0.1.3

[0.1.3] - 2026-04-20

Release automation bootstrap. No user-facing SDK changes.

Changed

• Added an automated release workflow (.github/workflows/release.yml) that bumps the version, generates the CHANGELOG.md entry via Gemini, publishes to GitHub Packages, and creates the GitHub Release behind an approval gate.
• Fixed Maven authentication in .github/workflows/publish.yml so manual publishes to GitHub Packages work; removed its release: published trigger to avoid duplicate deploys.