megachangelog
Fix0.0.73

SDK v0.0.73

Fixed stale managed Hub daemon reconnections by adding runtime build fingerprints, fixed compaction being skipped on reasoning models with improved summarizer token handling, added Fable 5 to Vertex model catalog, and improved custom Vertex model ID routing.

  • Fixed hosts reconnecting to stale managed Hub daemons: daemons now carry a runtime build fingerprint, so upgrading retires and respawns a daemon still running older code instead of attaching to it
  • Fixed compaction being silently skipped on reasoning models. The summarizer no longer hardcodes a 1024-token output cap — it honors an explicit max-output-tokens setting, defaults to 4096 (lowered when the model reports less), and logs a diagnostic when a summary comes back empty
  • Added Fable 5 (claude-fable-5) to the Vertex model catalog. Pricing is intentionally omitted because Vertex bills region-dependently, so cost shows as unknown rather than wrong
  • Custom Vertex model IDs are now passed through unchanged, routing Claude-style IDs to the Anthropic-on-Vertex path

Full Changelog: sdk/sdk/v0.0.72...sdk/sdk/v0.0.73

sdkdaemonmodelsvertexreliability

Source: original entry ↗