Multi-CLI Selection Became Real

Saved coding-CLI choices now flow through execution end to end, and the agent workspace does a better job exposing which providers are actually available.

This release moved multi-CLI support from promise to reality. The saved execution preference on an agent now matters during real work, and the workspace got better at showing which tools and credentials are actually available.

Saved Execution Choices Now Drive Real Runs

The inner-loop selector stopped being mostly cosmetic.

  • An agent's saved coding CLI now flows through the execution chain
  • Inner-loop settings load faster and fail less often, which makes the profile view feel less fragile
  • The agent profile became more execution-oriented, with less unrelated clutter around the settings that actually change how work runs

Credential Visibility Widened Across CLIs

The workspace now looks across the broader tool fleet instead of treating one provider as the entire story.

  • Claude, Codex, Gemini, and OpenCode are all surfaced in the auth view
  • Saved CLI and model choices rehydrate more reliably in the UI
  • Credential checks lean on lighter probes, so the page spends less time doing expensive work just to show status