Sep 25, 2026

New Models, Grouped PR Checks, and a Test Pruner Template

#GPT-6 Sol, GPT-6 Luna, and Claude Opus 5.5

Three new models join the lineup: GPT-6 Sol and GPT-6 Luna extend the GPT-6 family alongside Astra, and Claude Opus 5.5 is now available across every supported agent harness. Pick any of them from the model selector like usual.

#Pull request checks are grouped and update live

The task page now groups CI checks together instead of listing them one by one, and their status updates live on GitHub, GitLab, and Bitbucket as checks run, so you can tell whether a pull request is ready to merge without refreshing.

#Run agents when a GitHub Action finishes

Agents can now trigger off a GitHub Actions workflow run completing, filtered by repository, workflow, branch, or outcome, so you can kick off a follow-up the moment a build fails or passes.

#A Test Pruner agent template

The new Test Pruner template scans your test suite for redundant or low-value tests and proposes what to remove, ready to run straight from the template gallery.

#Everyday improvements

  • Grok is upgraded from 4.6 to 4.7.
  • Supabase now connects through the MCP catalog instead of the old OAuth flow.
  • Public MCP now supports OAuth and bearer-token authentication, making it easier to connect external MCP clients.
  • MCP servers that authenticate with OAuth Client ID Metadata Documents (CIMD) are now supported.
  • Self-hosted deployments can now run the web control plane from a Docker image.

Also various bug fixes and improvements as always :)