Skip to content

fix(claude-code): honor configured models - #2967

Open
r266-tech wants to merge 1 commit into
vectorize-io:mainfrom
r266-tech:fix/claude-code-forward-configured-model-2881
Open

fix(claude-code): honor configured models#2967
r266-tech wants to merge 1 commit into
vectorize-io:mainfrom
r266-tech:fix/claude-code-forward-configured-model-2881

Conversation

@r266-tech

Copy link
Copy Markdown
Contributor

Summary

  • pass the resolved Claude Code model to both SDK option paths
  • preserve empty-model behavior by omitting the SDK keyword
  • cover standard and tool-enabled paths, including the installed SDK contract

Fixes #2881.

This also makes the existing provider-aware default effective when no per-scope
model is configured; it does not change the default-model table.

Verification

  • uv run pytest tests/test_claude_code_llm_isolation.py -q (7 passed)
  • uv run pytest tests/test_provider_default_models.py tests/test_per_operation_llm_config.py -q
    (16 passed; one unrelated test could not start embedded Postgres because the
    sparse test environment lacks the optional pg0-embedded package)
  • targeted Ruff check and format check
  • targeted ty check
  • git diff --check
  • Codex adversarial review: approved

Thanks @bchumerch for the production reproduction and local validation, and
@ebarkhordar for verifying the current-main call sites and SDK contract.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

claude-code provider never passes the configured model to the CLI — HINDSIGHT_*_LLM_MODEL silently ignored, all calls run the CLI's own default

1 participant