fix(opencode): guard reasoningSummary for azure in options()

[jeanibarz]

Issue for this PR

Closes #21237

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

ProviderTransform.options() sets reasoningSummary = "auto" for all gpt-5 models unconditionally. The textVerbosity parameter in the same block already has && input.model.providerID !== "azure", but reasoningSummary does not. This adds the same guard.

Scope note: ProviderTransform.variants() also sets reasoningSummary for Azure variants — that's left intentional. The existing test at line 2372 ("standard azure models return custom efforts with reasoningSummary") explicitly expects it, and the default Azure path uses sdk.responses() which tolerates the parameter. This PR only fixes the asymmetric default in options(), which fires unconditionally regardless of API path.

How did you verify your code works?

1. Wrote a verification script that imports the real ProviderTransform module (not mocks) and exercises options(), variants(), and providerOptions() with Azure model configs matching the repo's models-api.json fixtures. Confirmed:

   • options() no longer sets reasoningSummary for Azure gpt-5 models
   • variants() still does (intentional, tested by maintainers)
   • providerOptions() wrapping correctly reflects the change

2. Unit tests added and all 123 existing transform.test.ts tests pass:

   cd packages/opencode && bun test -- test/provider/transform.test.ts
   

3. Could not run the full repo-wide bun typecheck locally — it fails on pre-existing Buffer type errors in unrelated files (clipboard.ts, cross-spawn-spawner.ts, db.node.ts, filesystem.ts). These same errors exist on a clean checkout of upstream/dev with bun@1.3.11. CI should pass as it uses a different environment.

Screenshots / recordings

N/A

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

Co-Authored-By: Claude Opus 4.6 (1M context) noreply@anthropic.com

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

The search results show PR #21575 (the current PR itself) and two potentially related historical PRs that dealt with reasoningSummary issues, but neither are duplicates of the current PR:

• PR fix: OpenCode was injecting reasoningSummary for GPT-5 and forwarding it through the openai-compatible chat adapter #14783: "fix: OpenCode was injecting reasoningSummary for GPT-5 and forwarding it through the openai-compatible chat adapter" — This is a historical fix about a different issue (injecting and forwarding the parameter), not about guarding it conditionally for Azure in options().

• PR feat(opencode): Add Databricks provider support #7984: "feat(opencode): Add Databricks provider support" — Unrelated feature PR.

No duplicate PRs found