feat(cli): honor --model alone without requiring --provider

[0xfandom]

Summary

• --model <name> now works standalone (no --provider required), overriding the active session model process-scoped.
• Routes the value to the env var matching the already-active provider (OPENAI_MODEL / GEMINI_MODEL / MISTRAL_MODEL / ANTHROPIC_MODEL) after saved-profile application and before the startup banner, so the banner, provider resolution, and outgoing request payload all agree.
• Preserves the existing --provider <name> --model <name> path (handled by applyProviderFlagFromArgs).

Closes #808.

Why

Today Commander parses --model inside main.tsx, but the startup banner and provider-env-reading code run earlier. Saved-profile users running openclaude --model gpt-4o-mini saw their stale profile model in the banner, and callers reading the *_MODEL env var directly (e.g. StartupScreen.detectProvider) missed the override entirely.

The fix mirrors the existing --provider early-path in cli.tsx: hook into the bootstrap flow right before printStartupScreen() and propagate the flag value to the correct env var.

No writes to .openclaude-profile.json — override is process-scoped, as the issue asked.

Changes

• src/utils/providerFlag.ts — export parseModelFlag, add applyModelFlagFromArgs that detects active provider from CLAUDE_CODE_USE_* and sets the matching *_MODEL.
• src/entrypoints/cli.tsx — call applyModelFlagFromArgs after saved-profile env application and before the startup banner, gated on --model being present.
• src/utils/providerFlag.test.ts — new tests covering parseModelFlag and applyModelFlagFromArgs across OpenAI / Gemini / Mistral / GitHub / Anthropic providers, saved-profile override, colon-bearing Ollama tags, and the --provider passthrough no-op.

Test plan

• bun test src/utils/providerFlag.test.ts — 34 pass, 0 fail
• bun test (full suite) — 1227 pass, 0 fail
• Saved OpenAI profile + openclaude --model gpt-4o-mini → banner shows gpt-4o-mini, profile file unchanged
• Saved Gemini profile + openclaude --model gemini-2.0-flash → banner shows gemini-2.0-flash
• openclaude --provider openai --model gpt-4o → existing behavior preserved

[gnanam1990]

Thanks! Closes #808 cleanly — defers to applyProviderFlagFromArgs when both flags are set, and the 9-test matrix covers every provider branch plus the ollama colon-syntax case. LGTM 🚀

[0xfandom]

@kevincodex1 Any update on this

[gnanam1990]

Re-approved at 1fc3ccc after the rebase onto current main.

Verified locally:

bun test src/utils/providerFlag.test.ts src/utils/providerAutoDetect.test.ts
→ 64 pass / 0 fail

The --model resolution still defers cleanly to applyProviderFlagFromArgs when both flags are present, and the matrix (9 provider branches + the ollama colon-syntax case) still passes after the rebase. No fingerprints / red flags introduced. LGTM 🚀

[0xfandom]

Rebased onto current main at ddbbee9 to clear the 7-commit drift from #849 / #972 / #983 / #979 / #980. No file overlap with any of those — only src/entrypoints/cli.tsx, src/utils/providerFlag.ts, src/utils/providerFlag.test.ts change here, none of which the new commits touch. Same diff as before (138 insertions, 1 deletion); bun run build + bun test src/utils/providerFlag.test.ts (49/49) green.

@gnanam1990 — re-approval ping if it still looks clean.

[0xfandom]

Rebased onto current main (post 0.9.2 release). 17-commit drift cleared with no conflicts — only src/entrypoints/cli.tsx + src/utils/providerFlag.{ts,test.ts} changed; nothing in those merged commits touches them.

bun test src/utils/providerFlag.test.ts src/utils/providerAutoDetect.test.ts
→ 75 pass / 0 fail
bun run build → green

No logic changes since the prior approval.

[0xfandom]

Any update on this @kevincodex1 @gnanam1990