feat: improve config provider TUI interaction

[5wxwxwxw5]

Description

• Support wrap-around navigation in provider, custom provider, and model lists (up at first item jumps to last, down at last item jumps to first)
• Add d key to delete custom models in model selection step with confirmation prompt
• Add protocol selection (anthropic/openai) to manual configuration flow and respect UseAnthropic setting in saved config
• Preserve manual protocol selection when re-entering the form
• Fix manual form input fields losing focus after returning from next step
• Use safe slice removal in removeFromSlice to avoid backing array corruption

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Refactoring (no functional changes)
• Documentation update
• CI / Build / Tooling

How Has This Been Tested?

• make test passes locally
• Manual testing (describe below)

Checklist

• My code follows the project's coding style (go fmt, go vet)
• I have performed a self-review of my code
• I have added tests that prove my fix is effective or my feature works
• New and existing unit tests pass locally with my changes
• I have updated the documentation accordingly (if applicable)
• I have signed the CLA

Related Issues

[github-actions]

🔍 OpenCodeReview found 4 issue(s) in this PR.

• ✅ 4 posted as inline comment(s)
• 📝 0 posted as summary

[github-actions]

Bug: When transitioning from manualStepURL to manualStepProtocol, the URL input is never blurred. The old code called m.manualURLInput.Blur() before advancing, but this was removed. Since the protocol step has no input widget and focusManualStep() returns nil for it, the URL text input retains focus state. This can cause visual artifacts (e.g., cursor still blinking in the URL field) while the protocol selector is displayed.

Similarly, the transition from manualStepModel to manualStepAuthToken also lost its m.manualModelInput.Blur() call.

You should blur the current step's input before advancing to the next step.

Suggestion:

Suggested change

	m.manualStep = manualStepProtocol
	return m, nil
	case manualStepProtocol:
	m.manualStep = manualStepModel
	return m, m.manualModelInput.Focus()
	m.manualURLInput.Blur()
	m.manualStep = manualStepProtocol
	return m, nil
	case manualStepProtocol:
	m.manualStep = manualStepModel
	return m, m.manualModelInput.Focus()

[github-actions]

Same issue: the model input should be blurred before transitioning to the auth token step, otherwise the model input may retain visual focus artifacts.

Suggestion:

Suggested change

	m.manualStep = manualStepAuthToken
	return m, m.manualTokenInput.Focus()
	m.manualModelInput.Blur()
	m.manualStep = manualStepAuthToken
	return m, m.manualTokenInput.Focus()

[github-actions]

Potential issue: If len(m.providers) is 0 (e.g., llm.ListProviders() returns empty), then len(m.providers) - 1 evaluates to -1, setting officialIdx to an invalid negative index. While unlikely in practice, adding a guard would make this more robust. The same applies to customListCount() and modelCount() which always return at least 1 due to the +1 sentinel item, so those are safe.

Suggestion:

Suggested change

	} else {
	m.officialIdx = len(m.providers) - 1
	}
	} else if len(m.providers) > 0 {
	m.officialIdx = len(m.providers) - 1
	}

[github-actions]

Minor readability issue: The condition i == m.manualProtocolIdx is checked twice in succession — once to set the cursor prefix and again to choose the rendering style. These can be consolidated into a single if/else block for clarity.

Suggestion:

Suggested change

	for i, proto := range cpProtocols {
	cur := " "
	if i == m.manualProtocolIdx {
	cur = " " + tuiCursorStyle.Render(tuiCursor) + " "
	}
	if i == m.manualProtocolIdx {
	s.WriteString(cur + tuiSelectedItemStyle.Render(proto) + "\n")
	} else {
	s.WriteString(cur + tuiItemStyle.Render(proto) + "\n")
	}
	}
	for i, proto := range cpProtocols {
	if i == m.manualProtocolIdx {
	cur := " " + tuiCursorStyle.Render(tuiCursor) + " "
	s.WriteString(cur + tuiSelectedItemStyle.Render(proto) + "\n")
	} else {
	s.WriteString(" " + tuiItemStyle.Render(proto) + "\n")
	}
	}

[lizhengfeng101]

Code Review Findings

Hi! I spotted a few things while reviewing — please take a look and let me know if you agree they're worth addressing.

1. Indentation / brace alignment in updateDeleteModelConfirm (potential correctness risk)

The indentation in this function is inconsistent, making it hard to verify the brace matching is correct:

			m.savedInSession = true
		if len(m.models()) > 0 {          // ← indent level drops unexpectedly
			if remaining := len(m.models()); m.modelIdx >= remaining {
				...
			}
		}
	}
		m.confirmingDeleteModel = false    // ← indent inconsistent with surrounding code
		return m, nil

Go uses braces (not indentation) for scoping, so this may compile fine — but the formatting makes it very difficult to audit correctness. Could you run gofmt on this function and verify the logic matches your intent?

2. Misplaced doc comment above cloneProviderEntry

The comment block describing applyEditCustomProviderSave sits directly above cloneProviderEntry:

// applyEditCustomProviderSave persists the edited provider to disk. Returns
// true if a hard validation error stopped the save ...
func cloneProviderEntry(v ProviderEntry) ProviderEntry {

Looks like the comment should be moved to the actual applyEditCustomProviderSave method.

3. removeFromSlice appears to be dead code

The PR description mentions "Use safe slice removal in removeFromSlice to avoid backing array corruption," but removeFromSlice is defined and never called anywhere — model deletion uses removeModels() instead. Should this function be removed, or is there a missing call site?

4. Redundant provider deletion (double save)

In updateDeleteConfirm, a deleted provider is both saved to disk immediately (via saveConfig) and appended to deletedProviders. Then in the confirmed path of runConfigProvider, applyProviderDeletions attempts to delete and save again. Since delete(map, key) is idempotent this isn't a bug, but the double-write adds unnecessary complexity. Consider either skipping the deletedProviders append when saving in-session, or guarding the post-TUI path.

5. m.models() called multiple times in updateDeleteModelConfirm

m.models() (which includes sorting) is called 3 times in sequence:

if len(m.models()) > 0 {
    if remaining := len(m.models()); m.modelIdx >= remaining {

Caching the result in a local variable would be cleaner and avoids redundant work:

updated := m.models()
if len(updated) > 0 && m.modelIdx >= len(updated) {
    m.modelIdx = len(updated) - 1
}