fix(agents): refresh curated model catalogs

Merged
KI
1mo ago
+185-1

What & why

Agent model pickers lagged behind provider releases. Add Opus 5.5 and Sonnet 5 to OpenCode/OMP, current Claude and GPT-6/GPT-5.6 choices to Copilot, newer Gemini choices, and Opus 5.5/Grok 4.7/Muse Spark 1.3 to Cursor with their supported effort variants. Add Sonnet 5 to the cloud chat catalog. Claude and Codex already had their current flagship models.

Preserve existing saved choices and configured defaults. Use each CLI's exact identifiers, including Copilot's dotted Claude versions and Cursor's effort-specific model IDs. docs/agent-model-catalog.md records sources and live-test results.

How I tested it

  • 165 targeted catalog, launch, settings, and host-validation tests passed.
  • Shared-package typecheck, changed-file Biome checks, strict i18n checks, and plugin checks passed.
  • Repository-wide bun run lint and bun run typecheck passed (42 typecheck tasks).
  • Full bun run test failed in the desktop suite: 7,651 passed, 34 failed, primarily git/worktree and shell-environment tests outside the changed files (including git-task-handlers.test.ts, external-worktree import, shell environment, and worktree hook tolerance). The full suite is not green; baseline failures were not independently reproduced on main.
  • All Cursor picker IDs and effort variants matched its live catalog; Codex's live model/list confirmed its model/effort choices.
  • Live smoke tests covered 33 configurations: Claude Opus 5.5/Sonnet 5 and Codex GPT-6 Astra/Sol/Luna returned the expected response. Copilot's 17 choices were blocked by account policy, Cursor's 3 by its free plan, Gemini's 2 by unsupported-client eligibility, and OpenCode's 4 by provider availability or missing credentials. OMP's 2 were not run because the CLI is not installed. No account policies or credentials were changed.
  • No desktop UI screenshot or end-to-end chat UI verification; live tests exercised the agent CLIs directly.

Checklist

  • PR title follows conventional commits (type(scope): subject)
  • bun run lint and bun run typecheck pass (CI fails on lint warnings too)
  • "Allow edits from maintainers" is checked on fork PRs (not applicable: same-repository branch)