- Release the post-v0.3.0 JSON-lines / Content-Length stdio framing compat for
--allow-startintrospection.allow-start-demonow prints
STDIO_FRAMING_*=PASSproof lines without changing the transport
implementation. - Parallelize Ollama embeddings in
semantic-selectvia bounded stdlib
ThreadPoolExecutorbatching when--embedding-backend ollamais selected.
prove-parallel-ollama-demoprintsPARALLEL_OLLAMA_BATCHED=true;
semantic-demowith fixture backend printsSEMANTIC_SELECT_STATUS=PASS. - Defer live-response-compression and broader CLI polish to v0.5.
- Post-review-pipeline hardening folded into this release: fast fail-closed
shutdown (cancel_futures) + bounded in-flight window for the parallel Ollama
path, fixture-onlysemantic-demo, matched stdio-framing fixture mode, and an
observable (non-traceback)prove_stdio_framingfailure path.
Full changelog: see CHANGELOG.md 0.4.0.