Package v0.2.12→v0.2.124: manifest, actions, version graph

startos/versions/v0.2.35.ts

@@ -0,0 +1,13 @@
+import { VersionInfo } from '@start9labs/start-sdk'
+
+export const v_0_2_35 = VersionInfo.of({
+  version: '0.2.35:0',
+  releaseNotes: {
+    en_US:
+      'Mirrors the Recap-app v0.2.68 transcribe-truncation fixes into the relay\'s server-side pipeline. (1) Relay\'s Gemini transcribe backend now sets maxOutputTokens: 65536 on its generateContent calls. Same root cause as the Recap-app fix — Gemini\'s default cap (~8192) silently truncates long-chunk transcripts mid-stream, costing the user 14-43 minutes of speech per affected chunk with no visible error. The relay\'s test-run benchmark suite was experiencing the same bug but the symptom got masked by the bigger chunks-share-TX optimization. (2) Gap-handling added to the relay\'s chunked-analyze.js planAnalysisWindows. When the transcript has a hole (e.g., one chunk truncated by Gemini\'s output cap), the old planner stopped at the hole and silently dropped any entries past it from analysis. New behavior: jump the body cursor forward to the next entry\'s body-stride boundary and keep planning windows instead of stopping. The /relay/summarize-url unified pipeline (and the test-run worker) both inherit this fix.',
+  },
+  migrations: {
+    up: async ({ effects }) => {},
+    down: async ({ effects }) => {},
+  },
+})