fix: make initial model selection respect active provider env by JithendraNara · Pull Request #306 · Gitlawb/openclaude

JithendraNara · 2026-04-04T02:16:10Z

Summary

make getUserSpecifiedModelSetting() choose the model env var for the active provider instead of checking unrelated provider env vars first
add regression tests for stale GEMINI_MODEL / OPENAI_MODEL cross-provider leakage
include those model-selection tests in the existing provider test suite

Impact

user-facing impact: OpenAI-compatible sessions no longer pick a stale Gemini model when both OPENAI_MODEL and GEMINI_MODEL are present in the environment
developer/maintainer impact: provider env precedence is now covered by the normal provider test path, reducing the chance of future regressions

Root Cause

getUserSpecifiedModelSetting() in src/utils/model/model.ts previously resolved model env vars in a fixed order:

ANTHROPIC_MODEL
GEMINI_MODEL
OPENAI_MODEL

That ignored the active provider and allowed stale env vars from a different provider to win.

A concrete failure case was:

CLAUDE_CODE_USE_OPENAI=1
OPENAI_MODEL=llama-3.3-70b-versatile
stale GEMINI_MODEL=gemini-2.0-flash-exp

In that state, the runtime could still resolve gemini-2.0-flash-exp even though the UI/provider path was OpenAI-compatible.

Fix

Make model selection provider-aware:

Gemini provider reads GEMINI_MODEL
OpenAI/Codex/GitHub providers read OPENAI_MODEL
first-party Anthropic reads ANTHROPIC_MODEL

This keeps model resolution aligned with the active provider instead of unrelated leftover env vars.

Testing

bun test src/utils/model/modelSelection.test.ts
env -u USER_TYPE bun run test:provider
env -u USER_TYPE bun run smoke
env -u USER_TYPE npm run test:provider-recommendation

Notes

provider/model path tested: OpenAI-compatible, Gemini, and first-party env precedence through the model-selection regression tests
screenshots attached: n/a
follow-up work or known limitations: there is a separate interactive startup issue where the splash screen can appear without the REPL prompt in some directories; that is intentionally not included in this PR to keep scope focused

Relates to #300

Copilot

Pull request overview

This PR fixes cross-provider model selection by making getUserSpecifiedModelSetting() read the model environment variable that corresponds to the active API provider, preventing stale model env vars from unrelated providers from taking precedence (relates to issue #300).

Changes:

Update getUserSpecifiedModelSetting() to choose GEMINI_MODEL / OPENAI_MODEL / ANTHROPIC_MODEL based on getAPIProvider().
Add regression tests covering stale GEMINI_MODEL/OPENAI_MODEL leakage across providers.
Include model-related tests in the existing test:provider script.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
src/utils/model/modelSelection.test.ts	Adds regression tests to ensure provider-aware model env precedence.
src/utils/model/model.ts	Makes env var model resolution provider-aware instead of fixed-order.
package.json	Expands `test:provider` to include `src/utils/model/*.test.ts`.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-04T02:20:39Z

+function clearProviderAndModelEnv(): void {
+  delete process.env.ANTHROPIC_MODEL
+  delete process.env.GEMINI_MODEL
+  delete process.env.OPENAI_MODEL
+  delete process.env.CLAUDE_CODE_USE_GEMINI
+  delete process.env.CLAUDE_CODE_USE_GITHUB
+  delete process.env.CLAUDE_CODE_USE_OPENAI
+}


clearProviderAndModelEnv() only clears Gemini/GitHub/OpenAI provider flags. If CLAUDE_CODE_USE_BEDROCK, CLAUDE_CODE_USE_VERTEX, or CLAUDE_CODE_USE_FOUNDRY are set in the runner environment, the "first-party" test may not actually execute the firstParty branch. Clear the remaining provider-selection env vars here (consistent with src/utils/model/providers.test.ts).

Copilot · 2026-04-04T02:20:39Z

+test('openai provider ignores stale gemini model env when choosing the main loop model', () => {
+  clearProviderAndModelEnv()
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_MODEL = 'llama-3.3-70b-versatile'
+  process.env.GEMINI_MODEL = 'gemini-2.0-flash-exp'
+
+  expect(getUserSpecifiedModelSetting()).toBe('llama-3.3-70b-versatile')
+})
+
+test('gemini provider ignores stale openai model env when choosing the main loop model', () => {
+  clearProviderAndModelEnv()
+  process.env.CLAUDE_CODE_USE_GEMINI = '1'
+  process.env.GEMINI_MODEL = 'gemini-2.5-flash'
+  process.env.OPENAI_MODEL = 'gpt-4o'
+
+  expect(getUserSpecifiedModelSetting()).toBe('gemini-2.5-flash')
+})


These test names say "choosing the main loop model", but the assertion is on getUserSpecifiedModelSetting() (a pre-allowlist/pre-default selection step). Consider renaming the tests (or asserting on getMainLoopModel() if that's what you intend) to keep the test intent precise.

JithendraNara · 2026-04-04T02:27:34Z

Addressed the Copilot review points in the latest push.

Expanded clearProviderAndModelEnv() to also clear CLAUDE_CODE_USE_BEDROCK, CLAUDE_CODE_USE_VERTEX, and CLAUDE_CODE_USE_FOUNDRY, so the first-party test stays isolated even if those flags are present in the runner environment.
Renamed the regression tests to describe the exact function under test: getUserSpecifiedModelSetting().

Re-ran:

env -u USER_TYPE bun test src/utils/model/modelSelection.test.ts
env -u USER_TYPE bun run test:provider

kevincodex1 · 2026-04-04T02:44:08Z

+  CLAUDE_CODE_USE_BEDROCK: process.env.CLAUDE_CODE_USE_BEDROCK,
+  CLAUDE_CODE_USE_FOUNDRY: process.env.CLAUDE_CODE_USE_FOUNDRY,


we are moving away from CLAUDE can we rename this constants to OPEN_CLAUDE ?

Vasanthdev2004

Looks good to merge from my side.

This targets the right root cause for #300: model selection now follows the active provider instead of letting stale env vars from another provider win. The new regression coverage around OPENAI_MODEL vs GEMINI_MODEL is solid, and pulling the model-selection tests into the provider test path makes sense.

I reran focused validation here with:

bun test ./src/utils/model/modelSelection.test.ts
bun test ./src/services/api/codexShim.test.ts ./src/utils/model/modelSelection.test.ts ./src/utils/model/providers.test.ts
cmd /c "set USER_TYPE=& bun run smoke"
cmd /c "set USER_TYPE=& npm run test:provider-recommendation"

kevincodex1 · 2026-04-04T09:23:27Z

please fix conflicts

JithendraNara · 2026-04-09T03:06:07Z

Rebased onto the latest main and resolved the merge conflicts.

Re-ran on the rebased branch:

env -u USER_TYPE bun test src/utils/model/modelSelection.test.ts
env -u USER_TYPE bun run test:provider
env -u USER_TYPE bun run smoke

GitHub now reports the PR as mergeable again; waiting on review/checks to refresh on the new head commit.

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-09T03:09:06Z

    // Read the model env var that matches the active provider to prevent
    // cross-provider leaks (e.g. ANTHROPIC_MODEL sent to the OpenAI API).
    const provider = getAPIProvider()
    specifiedModel =
      (provider === 'gemini' ? process.env.GEMINI_MODEL : undefined) ||
-      (provider === 'openai' || provider === 'gemini' || provider === 'github'
+      (provider === 'openai' || provider === 'codex' || provider === 'github'
        ? process.env.OPENAI_MODEL
        : undefined) ||
-      (provider === 'firstParty' ? process.env.ANTHROPIC_MODEL : undefined) ||
+      process.env.ANTHROPIC_MODEL ||
      settings.model ||


getUserSpecifiedModelSetting() currently falls back to process.env.ANTHROPIC_MODEL unconditionally. That can still cause cross-provider leakage when the active provider is openai/codex/github/gemini and the provider-specific model env var is unset, contradicting the inline comment about preventing ANTHROPIC_MODEL being sent to the OpenAI API. Consider only reading ANTHROPIC_MODEL when the active provider is Anthropic-backed (e.g. firstParty and any other providers that legitimately use ANTHROPIC_MODEL), and otherwise skip it so the function falls back to settings/defaults.

Copilot · 2026-04-09T03:09:06Z

+test('getUserSpecifiedModelSetting prefers OPENAI_MODEL for openai provider over stale GEMINI_MODEL', () => {
+  clearProviderAndModelEnv()
+  process.env.CLAUDE_CODE_USE_OPENAI = '1'
+  process.env.OPENAI_MODEL = 'llama-3.3-70b-versatile'
+  process.env.GEMINI_MODEL = 'gemini-2.0-flash-exp'
+
+  expect(getUserSpecifiedModelSetting()).toBe('llama-3.3-70b-versatile')
+})


The getUserSpecifiedModelSetting() logic now treats codex as an OpenAI-compatible provider (reads OPENAI_MODEL), but the regression tests here only cover openai, gemini, and firstParty. Adding a codex case (e.g., CLAUDE_CODE_USE_OPENAI=1 with a Codex-detected OPENAI_MODEL) would lock in the intended behavior and prevent future regressions for that provider branch.

Copilot AI review requested due to automatic review settings April 4, 2026 02:16

Copilot started reviewing on behalf of JithendraNara April 4, 2026 02:16 View session

Copilot AI reviewed Apr 4, 2026

View reviewed changes

kevincodex1 requested changes Apr 4, 2026

View reviewed changes

Vasanthdev2004 previously approved these changes Apr 4, 2026

View reviewed changes

JithendraNara added 3 commits April 8, 2026 23:04

fix: make initial model selection respect active provider env

72d7a5c

test: harden provider model env regression coverage

2cbfa0c

test: rename local env snapshots in model selection tests

edeec0c

Copilot AI review requested due to automatic review settings April 9, 2026 03:05

JithendraNara dismissed Vasanthdev2004’s stale review via edeec0c April 9, 2026 03:05

JithendraNara force-pushed the codex/fix-provider-model-env branch from e1e78a0 to edeec0c Compare April 9, 2026 03:05

Copilot started reviewing on behalf of JithendraNara April 9, 2026 03:05 View session

Copilot AI reviewed Apr 9, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: make initial model selection respect active provider env#306

fix: make initial model selection respect active provider env#306
JithendraNara wants to merge 3 commits intoGitlawb:mainfrom
JithendraNara:codex/fix-provider-model-env

JithendraNara commented Apr 4, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 4, 2026

Uh oh!

Copilot AI Apr 4, 2026

Uh oh!

JithendraNara commented Apr 4, 2026

Uh oh!

kevincodex1 Apr 4, 2026

Uh oh!

Vasanthdev2004 left a comment

Uh oh!

kevincodex1 commented Apr 4, 2026

Uh oh!

JithendraNara commented Apr 9, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 9, 2026

Uh oh!

Copilot AI Apr 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

		CLAUDE_CODE_USE_BEDROCK: process.env.CLAUDE_CODE_USE_BEDROCK,
		CLAUDE_CODE_USE_FOUNDRY: process.env.CLAUDE_CODE_USE_FOUNDRY,

Conversation

JithendraNara commented Apr 4, 2026

Summary

Impact

Root Cause

Fix

Testing

Notes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 4, 2026

Choose a reason for hiding this comment

Uh oh!

JithendraNara commented Apr 4, 2026

Uh oh!

kevincodex1 Apr 4, 2026

Choose a reason for hiding this comment

Uh oh!

Vasanthdev2004 left a comment

Choose a reason for hiding this comment

Uh oh!

kevincodex1 commented Apr 4, 2026

Uh oh!

JithendraNara commented Apr 9, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants