@pellux/goodvibes-tui 0.19.59 → 0.19.61

package/CHANGELOG.md

@@ -4,6 +4,58 @@ All notable changes to GoodVibes TUI.
 
 ---
 
+## [Unreleased]
+
+---
+
+## [0.19.61] — 2026-05-05
+
+### Changed
+- Updated `@pellux/goodvibes-sdk` to `0.28.22`.
+- Regenerated the operator contract artifacts for SDK 0.28.22.
+- Updated `@pellux/goodvibes-sdk` and all split GoodVibes packages to `0.33.4`.
+- Migrated TUI imports to SDK public package/subpath seams after the SDK removed private `_internal` and catch-all platform exports.
+- Regenerated foundation/operator contract artifacts for the SDK 0.33 public contract shape.
+
+### Verified
+- Verified the previously blocked `runtime/http-transport`, `runtime/realtime-transport`, and `release-gates/transport-parity` tests pass against SDK 0.33.4.
+- Verified the full intended TUI test suite passes without quarantines or skipped migration failures: 480 test files passed, 0 failed.
+- Rebuilt production TUI and daemon binaries with SDK 0.33.4 and smoke-tested the compiled daemon binary.
+- Restarted the compiled daemon against the existing `.goodvibes/tui/control-plane/sessions.json` control-plane store and verified authenticated `/status` reports SDK 0.33.4 without session-normalization startup errors.
+- Reran the TTS `STREAM_END` regression and verified provider-scoped `STREAM_END` does not end a spoken turn before the logical turn completes.
+- Rebuilt and relaunched the daemon with SDK 0.28.10, then retested Home Graph reindex coalescing, status responsiveness, TV feature Ask, base `knowledge.ask` with `knowledgeSpaceId: "homeassistant"`, generated pages, and map edge payloads against the live daemon.
+- Rebuilt and relaunched the daemon with SDK 0.28.11, verified reset `dryRun` is non-destructive against the live Home Graph space, and reran the full local test suite.
+- Rebuilt and relaunched the daemon with SDK 0.28.12, smoke-tested the compiled daemon binary, and coordinated the Home Assistant restart before live Home Graph validation.
+- Rebuilt and relaunched the daemon with SDK 0.28.13, smoke-tested the compiled daemon binary, and verified TTS spoken turns stay active through provider `STREAM_END` until the logical turn completes.
+- Live-tested SDK 0.28.13 Home Graph Ask/reindex/refinement: Ask now returns bounded current evidence with refinement task ids, and the coordinated Home Assistant follow-up showed background refinement can close the LG TV answer gaps; SDK follow-up is still needed for refinement-run budget failures, blocked retry-window tasks, and non-synthesized raw-snippet follow-up output.
+- Rebuilt and relaunched the daemon with SDK 0.28.14, smoke-tested the compiled daemon binary, and reran the TTS `STREAM_END` spoken-turn regression.
+- Coordinated the SDK 0.28.14 Home Assistant live pass: refinement runs now return immediately under budget, Ask/base Ask stay bounded, follow-up Ask synthesizes from repaired LG TV evidence, official LG sources appear in the repaired source set, and map/pages remain clean; one LG TV gap remains active with retryable/deferred repair state.
+- Rebuilt and relaunched the daemon with SDK 0.28.15, smoke-tested the compiled daemon binary, reran the TTS `STREAM_END` spoken-turn regression, and coordinated Home Assistant ownership of the live Home Graph route lane.
+- Coordinated the SDK 0.28.15 Home Assistant live pass: sync, reindex, refinement run, Ask, base Ask, and map stayed responsive; remaining SDK-side follow-up includes official-source answer preference, promoted typed facts, generated pages route reliability, and refinement limit accounting.
+- Rebuilt and relaunched the daemon with SDK 0.28.16, smoke-tested the compiled daemon binary, reran the TTS `STREAM_END` spoken-turn regression, and coordinated Home Assistant ownership of the live Home Graph route lane.
+- Coordinated the SDK 0.28.16 Home Assistant live pass: reset, sync, changed-only reindex, refinement run, pages, maps, and status checks stayed responsive; remaining SDK-side follow-up includes slow initial Home Graph Ask, weak first-answer synthesis with zero promoted facts, base knowledge Ask missing the official LG source, follow-up answers preferring secondary sources, and Home Graph map filters returning unchanged result counts.
+- Rebuilt and relaunched the daemon with SDK 0.28.17, smoke-tested the compiled daemon binary, reran the TTS `STREAM_END` spoken-turn regression, and ran the TUI-owned Home Graph evaluation before handing the route lane to Home Assistant.
+- Live-tested SDK 0.28.17 Home Graph reset/sync/reindex/refinement/Ask/pages/map behavior: destructive reset now deletes scoped generated artifacts, HA sync remains bounded, reindex coalesces overlapping calls, nested Home Assistant map filters narrow results, and base `knowledge.ask` now links concrete TV queries to the scoped LG Home Assistant device while leaving generic "the device" queries unlinked. Remaining SDK-side follow-up includes slow first Home Graph Ask, official LG source not driving the final/follow-up answer, top-level map filters being ignored, and generated page/follow-up prose still containing raw-snippet-style sections.
+- Rebuilt and relaunched the daemon with SDK 0.28.18, smoke-tested the compiled daemon binary, reran the TTS `STREAM_END` spoken-turn regression, and ran the TUI-owned Home Graph evaluation before handing the route lane to Home Assistant.
+- Live-tested SDK 0.28.18 Home Graph reset/sync/reindex/refinement/Ask/pages/map behavior: reset dry-run stayed non-destructive, destructive reset cleared scoped generated artifacts, HA sync returned in 5.4s, overlapping reindex coalesced, refinement/run stayed bounded, top-level and nested Home Assistant map filters both narrowed results with titled non-self-loop edges, and concrete TV asks linked only the real LG Home Assistant device while generic "the device" asks did not invent linked objects. Remaining SDK-side follow-up includes very slow first Home Graph Ask, base/follow-up answers still preferring secondary sources over the accepted official LG source, promoted facts missing subject/target metadata, generated pages missing subject/target/neighbors/relatedPages metadata and still rendering one raw-snippet-style LG page section, and generic base knowledge "device" asks returning noisy Home Assistant app facts.
+- Rebuilt and relaunched the daemon with SDK 0.28.19, smoke-tested the compiled daemon binary, reran the TTS `STREAM_END` spoken-turn regression, and ran the TUI-owned Home Graph evaluation before handing the route lane to Home Assistant.
+- Live-tested SDK 0.28.19 Home Graph reindex/refinement/Ask/pages/map behavior: changed-only reindex skipped missing generated-page artifacts without failures, broad refinement reported `requestedLimit: 500` and `effectiveLimit: 24`, Home Graph and base knowledge TV asks synthesized clean prose from official LG/vendor sources, concrete TV asks linked only the real LG Home Assistant device, returned facts now include top-level `subject`, `subjectIds`, `linkedObjectIds`, and `targetHints`, and generic "the device" asks return no unrelated Home Assistant app facts or gaps. Remaining SDK-side follow-up includes generated LG passport content still containing one raw/evidence-style section and page metadata still missing subject/target/neighbors/relatedPages fields.
+- Rebuilt and relaunched the daemon with SDK 0.28.20, smoke-tested the compiled daemon binary, reran the TTS `STREAM_END` spoken-turn regression, and ran the TUI-owned Home Graph evaluation before handing the route lane to Home Assistant.
+- Live-tested SDK 0.28.20 Home Graph reindex/refinement/Ask/pages/map behavior: overlapping reindex now exposes a coalesced bounded result, Home Graph and base knowledge TV asks continue to synthesize clean official/vendor-backed prose with only the real LG Home Assistant device linked, generic "the device" asks still return no unrelated facts or gaps, generated LG passports no longer match raw-evidence/direct-comparison/table-debris markers, and legitimate `2 x 10W`-style speaker specs remain present. Remaining SDK-side follow-up includes generated passport quality still containing low-value canonical fact text such as truncated marketing/spec fragments and page metadata still missing subject/target/neighbors/relatedPages fields.
+- Rebuilt and relaunched the daemon with SDK 0.28.21, smoke-tested the compiled daemon binary, reran the TTS `STREAM_END` spoken-turn regression, and ran the TUI-owned cold/warm Home Graph evaluation before handing the route lane to Home Assistant.
+- Live-tested SDK 0.28.21 Home Graph reset/sync/reindex/refinement/Ask/pages behavior: reset dry-run stayed non-destructive, destructive reset deleted 57 scoped artifacts, Home Assistant sync returned in 4.49s with status checks staying responsive, changed-only reindex finished in 592ms and skipped generated-page artifacts, concrete TV asks linked only the real LG Home Assistant device, generic "the device" asks returned no unrelated facts/gaps, warm Home Graph/base asks synthesized clean source-backed prose with 24 subject-linked facts, and `nextRepairAttemptAt` is present on blocked repair tasks. Remaining SDK-side follow-up includes slow cold asks, cold Home Graph Ask returning active repair metadata with 0 promoted facts before base Ask repairs the evidence, refinement/run exhausting its budget after 63.9s with `requestedLimit: 500` / `effectiveLimit: 24`, generated page metadata still missing subject/target/neighbors/relatedPages, and LG passport canonical facts still containing low-value truncated spec fragments despite no raw-evidence marker matches.
+- Rebuilt and relaunched the daemon with SDK 0.28.22, smoke-tested the compiled daemon binary, reran the TTS `STREAM_END` spoken-turn regression, and ran the TUI-owned focused warm Home Graph page-refresh validation before handing the route lane to Home Assistant.
+- Live-tested SDK 0.28.22 Home Graph Ask-to-page propagation: warm Home Graph Ask returned in 14.64s with status probes responsive, synthesized official/vendor-backed prose, linked only the real LG Home Assistant device, returned 12 subject-linked facts, and reported `pageRefreshRequested`/`pageRefreshed`; pages rendered 40 markdown pages in 103ms, the LG passport no longer shows `0 source(s)` or stale manual/source questions, preserves a `2 x 10W` speaker fact, and map sanity returned 95 titled non-self-loop edges in 27ms. Remaining SDK-side follow-up includes the LG passport still lacking explicit official LG source text, page metadata still missing subject/target/neighbors/relatedPages, and the passport still rendering low-value truncated canonical fact fragments such as HDMI quantity/table debris.
+
+---
+
+## [0.19.60] — 2026-05-01
+
+### Fixed
+- Embedded the `sql.js` WASM payload during the Bun compile compatibility pass so published daemon binaries no longer try to load `sql-wasm.wasm` from the GitHub runner's `node_modules` path at runtime.
+
+---
+
 ## [0.19.59] — 2026-05-01
 
 ### Changed

package/README.md

@@ -2,7 +2,7 @@
 
 [![CI](https://github.com/mgd34msu/goodvibes-tui/actions/workflows/ci.yml/badge.svg)](https://github.com/mgd34msu/goodvibes-tui/actions/workflows/ci.yml)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
-[![Version](https://img.shields.io/badge/version-0.19.59-blue.svg)](https://github.com/mgd34msu/goodvibes-tui)
+[![Version](https://img.shields.io/badge/version-0.19.61-blue.svg)](https://github.com/mgd34msu/goodvibes-tui)
 
 A terminal-native AI coding, operations, automation, knowledge, and integration console with a typed runtime, omnichannel surfaces, structured memory/knowledge, and a raw ANSI renderer.