fix(git): use user-selected model for auto-generated commit messages

[seifalmotaz]

What Changed

This PR fixes the auto-generated commit messages, PR content, and branch names to use the model selected by the user in the chat UI model picker, instead of the hardcoded gpt-5.3-codex constant.

Changes:

1. Contracts (packages/contracts/src/git.ts)

   • Added optional model field to GitRunStackedActionInput schema

2. Server - TextGeneration (apps/server/src/git/Services/TextGeneration.ts)

   • Added optional model parameter to CommitMessageGenerationInput, PrContentGenerationInput, and BranchNameGenerationInput interfaces

3. Server - CodexTextGeneration (apps/server/src/git/Layers/CodexTextGeneration.ts)

   • Renamed CODEX_MODEL to DEFAULT_CODEX_MODEL as fallback
   • Added model parameter to runCodexJson function
   • Uses provided model or falls back to default

4. Server - GitManager (apps/server/src/git/Layers/GitManager.ts)

   • Wires model parameter through runFeatureBranchStep, runCommitStep, runPrStep, and resolveCommitAndBranchSuggestion

5. Server - ProviderCommandReactor (apps/server/src/orchestration/Layers/ProviderCommandReactor.ts)

   • Passes model from thread.turn-start-requested event to branch name generation

6. Web - gitReactQuery (apps/web/src/lib/gitReactQuery.ts)

   • Added model parameter to gitRunStackedActionMutationOptions

7. Web - GitActionsControl (apps/web/src/components/GitActionsControl.tsx)

   • Added selectedModel prop and passes it to the mutation

8. Web - ChatHeader (apps/web/src/components/chat/ChatHeader.tsx)

   • Added selectedModel prop and passes it to GitActionsControl

9. Web - ChatView (apps/web/src/components/ChatView.tsx)

   • Supplies selectedModel to ChatHeader

Why

Previously, when a user selected a specific model (e.g., gpt-5.2-codex) in the chat UI model picker and then triggered a Git action like "Auto Commit" or "Commit & Push", the app would ignore that selection and use the hardcoded gpt-5.3-codex model. This was problematic because:

• Users may not have access to the hardcoded model
• The model selection should be respected consistently across all features
• It caused unexpected errors when the hardcoded model wasn't available

The fix propagates the user-selected model through the entire chain from the UI to the Codex CLI text generation calls, while maintaining backward compatibility with a sensible default fallback.

UI Changes

No visible UI changes. The behavior is now consistent with user expectations - the selected model is used for all AI-powered operations including commit message generation.

Checklist

• This PR is small and focused
• I explained what changed and why
• I included before/after screenshots for any UI changes (not applicable - no visible UI changes)
• I included a video for animation/interaction changes (not applicable)

Fixes #1165

Note

Pass user-selected model to auto-generated commit messages, branch names, and PR content

• Adds an optional model field to the GitRunStackedActionInput contract and threads it through GitManager into all text generation calls (generateCommitMessage, generateBranchName, generatePrContent).
• CodexTextGeneration now accepts a model override in each generation handler and passes it to the codex --model flag, falling back to DEFAULT_CODEX_MODEL when not supplied.
• The web UI reads the currently selected model from ChatView, passes it via ChatHeader to GitActionsControl, and includes it in the runImmediateGitActionMutation payload.
• First-turn branch name generation in ProviderCommandReactor also receives the model from the event payload.

^{Macroscope summarized 5a884a7.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 33bb314e-bda3-4b5f-b98f-0716e45803c3

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

📝 Coding Plan

• Generate coding plan for human review comments

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Tip

CodeRabbit can enforce grammar and style rules using `languagetool`.

Configure the reviews.tools.languagetool setting to enable/disable rules and categories. Refer to the LanguageTool Community to learn more.

[juliusmarminge]

i don't think this is the correct solution. i wouldn't wanna use gpt-5.4 xhigh to generate a commit message for me, way too slow and it will overthink it.

what we should fix:

• not using a model the user doesn't have access to (same as we do for gpt 5.3 spark)
• make it configurable (probably a knob in settings, we really need to do #1033 to have server authorative settings)

[seifalmotaz]

Thanks for the feedback! I understand now - my previous PR was wrong.
ou're right that using gpt-5.4-xhigh for commit messages would be overkill. The solution should:

. Sensible default: Use a fast, appropriate model (like gpt-5.3-codex) for commit messages, NOT the user's chat model selection
2. Configurable: Let users override this if they want
3. Access check: Fall back silently if user lacks access to the configured model
Question: Should I implement this as a client-side setting now (in localStorage), or is this blocked until #1033?
For the default, I'd use gpt-5.3-codex - sound good?

[juliusmarminge]

Thanks for the feedback! I understand now - my previous PR was wrong. ou're right that using gpt-5.4-xhigh for commit messages would be overkill. The solution should:

. Sensible default: Use a fast, appropriate model (like gpt-5.3-codex) for commit messages, NOT the user's chat model selection 2. Configurable: Let users override this if they want 3. Access check: Fall back silently if user lacks access to the configured model Question: Should I implement this as a client-side setting now (in localStorage), or is this blocked until #1033? For the default, I'd use gpt-5.3-codex - sound good?

can be client side for now yes.

default can probably be the newly released gpt-5.4-mini now

[seifalmotaz]

I will draft this PR as it's the wrong approach, and will create new PR with new approach, i will make it as soon as i test it and make sure it's in good place (very soon).
Thanks

[seifalmotaz]

#1171