feat(orchestrator): Page capability catalogs on demand#28

draft1
contrib/ctm/codex-background-batching
contrib/ctm/mcp-capability-pagination
openedSep 27, 2026byMMike Olson
View on GitHubTwo-way sync enabledPR last imported 3h ago

What Changed

orchestrator_capabilities called with no arguments now returns provider summaries, inherited settings and feature flags without any model catalogs. An agent passes providerInstanceId to read that provider's models in pages of 50 by default and 100 at most, continues with modelCursor, and asks for an exact model with includeModelOptions: true when it needs that model's option descriptors. Invalid combinations, unknown providers and unadvertised models return clear errors.

The tool description, the orchestration instructions given to agents and the MCP server docs describe the new flow.

Why

Providers with large catalogs made the default response huge, and agents spent context on every model and option descriptor just to find one provider ID. Paging lets them fetch only what they need.

Checklist

  • This PR is small and focused
  • I explained what changed and why
  • I included before/after screenshots for any UI changes
  • I included a video for animation/interaction changes

Added 1 commit

  • feat(orchestrator): Page capability catalogs on demand2bd98ec