Skip to content

v0.2.6

Choose a tag to compare

View all tags
@timofriedlberlin timofriedlberlin released this 20 Apr 00:52
· 40 commits to main since this release
v0.2.6
6c86e93

v0.2.6 — 2026-04-20

Improved model resolution visibility, request caching hints, and SWE-bench agent benchmarking ergonomics.

Added

  • Printed the provider-resolved model once per streamed step when the model becomes known
  • Added request-cache coverage in agent tests to ensure top-level cache hints are propagated from message history
  • Added a dedicated bench:swe:agent:go Task target for the live Go-centric SWE benchmark subset
  • Added live per-instance progress output in the SWE bench runner and optional live agent log streaming via SWE_LIVE_OUTPUT=1
  • Added tests covering non-overlapping output and reasoning token aggregation and display

Changed

  • Moved model resolution display into the agent streaming path instead of resolving model aliases eagerly in main
  • Updated github.com/codewandler/agentapis to v0.3.2
  • Extended the SWE bench Docker task setup to mount Codex credentials read-only for agent-driven runs

Fixed

  • Fixed agent tests to use valid llm test stream helpers
  • Fixed usage aggregation and display expectations so output and reasoning tokens are reported separately without double counting
  • Removed obsolete top-level model resolution tests tied to the previous eager-resolution path
Assets 2
Loading