Cline SDK 0.0.73: compaction fix on reasoning models, Fable 5 on Vertex
Bug-fix release for the Cline SDK. Context compaction was being silently skipped on reasoning models because the summarizer hardcoded a 1024-token output cap; it now honors a configured max-output-tokens. Vertex users get Fable 5 in the model catalog and pass-through of custom Vertex model IDs.
PUBLISHED2026-08-11
OBSERVED2026-08-11
AGE14d
SOURCES1
- Summarizer no longer hardcodes a 1024-token output cap: honors an explicit max-output-tokens setting, defaults to 4096, lowers it when the model reports less, and logs a diagnostic on an empty summary. Fixes compaction being silently skipped on reasoning models.
- Managed Hub daemons carry a runtime build fingerprint, so an upgrade retires and respawns a daemon running older code instead of a host reattaching to the stale one.
- Added Fable 5 (
claude-fable-5) to the Vertex model catalog. Cost shows as unknown — pricing is omitted deliberately because Vertex bills per region. - Custom Vertex model IDs pass through unchanged; Claude-style IDs route to the Anthropic-on-Vertex path.
COMMUNITY
No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.