Skip to content

Changelog

0.2.0-alpha.1 (2026-10-28)

Features

  • workbench: show the agent card in a readable form on the Agents page (#217) (f5a650c), closes #138

0.2.0-alpha (2026-10-23)

⚠ BREAKING CHANGES

  • sdk: the contract version is 0.9.0; agents declaring 0.8.x are reported as incompatible.

Features

  • agents: a model-backed orchestrator that routes from the manifest (#161) (021f470)
  • app: route the app through the conductor, with its own configuration (#63) (1ac8e53)
  • ci: code coverage for the unit tests, with Codecov reporting and badges (#90) (b0f611e)
  • harness: a standalone A2A server for deployments (#55) (#65) (9fa54fa)
  • harness: a test kit of its own, so harness tests leave workbench/tests (#62) (bf72812), closes #56
  • harness: find agents that start after the registry was built (#165) (e4465c3)
  • orchestrator: default to Claude Haiku 5.5 (#172) (7aa861c)
  • sdk: a model key stays with the agent that calls the model (#157) (a846b8f)
  • sdk: record an authenticated principal and the issuer that vouched (#181) (57650b3)
  • workbench: colour agent status chips by state (#136) (63a05d2)

Bug Fixes

  • workbench: report a missing profile, agent list or source list in one line (#61) (afaaa8c)

Documentation

  • adr: the app's entry point is the orchestrator conversation (#137) (eba0ec2)
  • explain the system architecture from the top down (#148) (cac515c)
  • record why the contract is written in LinkML (ADR-0020) (#44) (a366461)
  • separate the docs by what they describe (#58) (#64) (b55f849)
  • stop naming specific models in prose and examples (#174) (ab5c074)
  • workbench: record what RO-Crate is for (ADR-0021) (#48) (b748203)
  • workbench: walk one run through every step (#45) (ab18167)