feat: Add multi-provider model support (OpenAI/Google/Codex) to model system

[JeremyDev87]

Summary

Codingbuddy serves as a multi-AI rules MCP server for all AI assistants (Cursor, Claude Code, GitHub Copilot/Codex, Antigravity, Amazon Q, Kiro), but the model configuration system is currently hard-coded for Anthropic Claude only. This creates a poor experience for users of other AI providers.

Problem

1. model.constants.ts — Claude-only constants

File: apps/mcp-server/src/model/model.constants.ts

Currently only defines Claude models:

```ts
export const CLAUDE_OPUS_4 = 'claude-opus-4-20250514';
export const CLAUDE_SONNET_4 = 'claude-sonnet-4-20250514';
export const CLAUDE_HAIKU_35 = 'claude-haiku-3-5-20241022';
export const DEFAULT_MODEL = CLAUDE_SONNET_4;
```

No constants exist for OpenAI (gpt-4o, o3-mini, o1-mini) or Google (gemini-2.5-pro, gemini-2.5-flash) models.

---

2. model.resolver.ts — False "Unknown model" warnings for GPT/Gemini

File: apps/mcp-server/src/model/model.resolver.ts

```ts
export const KNOWN_MODEL_PREFIXES = [
'claude-opus-4',
'claude-sonnet-4',
'claude-sonnet-3',
'claude-haiku-3',
] as const;
```

When a user sets ai.defaultModel: "gpt-4o" in codingbuddy.config.json, resolveModel() returns a warning:

Unknown model ID: "gpt-4o". Known prefixes: claude-opus-4, claude-sonnet-4, ...

The model still works, but the warning is misleading for valid OpenAI/Google models.

---

3. model-prompt.ts — CLI init only offers Claude choices

File: apps/mcp-server/src/cli/init/prompts/model-prompt.ts

```ts
export function getModelChoices(): ModelChoice[] {
return [
{ name: 'Claude Sonnet 4 (Recommended)', value: CLAUDE_SONNET_4, ... },
{ name: 'Claude Opus 4', value: CLAUDE_OPUS_4, ... },
{ name: 'Claude Haiku 3.5 (Not recommended)', value: CLAUDE_HAIKU_35, ... },
];
}
```

During codingbuddy init, Codex/GPT/Gemini users see only Claude options with no way to select their preferred model interactively.

---

Proposed Solution

Step 1 — Expand model.constants.ts with multi-provider constants

Add constants for OpenAI and Google models alongside the existing Claude constants:

```ts
// Existing Claude constants (keep as-is)
export const CLAUDE_OPUS_4 = 'claude-opus-4-20250514';
export const CLAUDE_SONNET_4 = 'claude-sonnet-4-20250514';
export const CLAUDE_HAIKU_35 = 'claude-haiku-3-5-20241022';

// OpenAI models
export const GPT_4O = 'gpt-4o';
export const GPT_4O_MINI = 'gpt-4o-mini';
export const O3_MINI = 'o3-mini';
export const O1_MINI = 'o1-mini';

// Google models
export const GEMINI_25_PRO = 'gemini-2.5-pro';
export const GEMINI_25_FLASH = 'gemini-2.5-flash';

// Default remains Claude Sonnet (backward compatible)
export const DEFAULT_MODEL = CLAUDE_SONNET_4;
```

---

Step 2 — Expand KNOWN_MODEL_PREFIXES in model.resolver.ts

```ts
export const KNOWN_MODEL_PREFIXES = [
// Anthropic Claude
'claude-opus-4',
'claude-sonnet-4',
'claude-sonnet-3',
'claude-haiku-3',
// OpenAI
'gpt-4',
'gpt-3.5',
'o1-',
'o3-',
// Google
'gemini-',
] as const;
```

Also remove the Haiku-specific deprecation logic (currently tied to Claude), or make it provider-agnostic.

---

Step 3 — Update model-prompt.ts CLI with multi-provider choices

Group choices by provider so Codex/GPT users can select their model during codingbuddy init:

```ts
export function getModelChoices(): ModelChoice[] {
return [
// Anthropic (default)
{ name: 'Claude Sonnet 4 (Recommended)', value: CLAUDE_SONNET_4, description: 'Balanced performance and cost · Anthropic' },
{ name: 'Claude Opus 4', value: CLAUDE_OPUS_4, description: 'Most capable · Anthropic' },
// OpenAI
{ name: 'GPT-4o', value: GPT_4O, description: 'Flagship multimodal model · OpenAI' },
{ name: 'GPT-4o mini', value: GPT_4O_MINI, description: 'Fast and affordable · OpenAI' },
{ name: 'o3-mini', value: O3_MINI, description: 'Reasoning model · OpenAI' },
// Google
{ name: 'Gemini 2.5 Pro', value: GEMINI_25_PRO, description: 'Advanced reasoning · Google' },
{ name: 'Gemini 2.5 Flash', value: GEMINI_25_FLASH, description: 'Fast and efficient · Google' },
// Escape hatch
{ name: 'Other (enter manually)', value: 'custom', description: 'Any model ID not listed above' },
];
}
```

When __custom__ is selected, follow up with an input prompt to collect a free-text model ID.

---

Files to Change

File	Change
apps/mcp-server/src/model/model.constants.ts	Add OpenAI and Google model constants
apps/mcp-server/src/model/model.constants.spec.ts	Add tests for new constants
apps/mcp-server/src/model/model.resolver.ts	Expand KNOWN_MODEL_PREFIXES
apps/mcp-server/src/model/model.resolver.spec.ts	Add tests for GPT/Gemini prefix recognition
apps/mcp-server/src/cli/init/prompts/model-prompt.ts	Multi-provider choices + __custom__ handler
apps/mcp-server/src/cli/init/prompts/model-prompt.spec.ts	Update/add tests for new choices

---

Acceptance Criteria

model.constants.ts

• GPT_4O, GPT_4O_MINI, O3_MINI, O1_MINI constants are exported
• GEMINI_25_PRO, GEMINI_25_FLASH constants are exported
• DEFAULT_MODEL remains claude-sonnet-4-20250514 (backward compatible)

model.resolver.ts

• isKnownModel('gpt-4o') returns true
• isKnownModel('gpt-4o-mini') returns true
• isKnownModel('o3-mini') returns true
• isKnownModel('o1-mini') returns true
• isKnownModel('gemini-2.5-pro') returns true
• isKnownModel('gemini-2.5-flash') returns true
• resolveModel({ globalDefaultModel: 'gpt-4o' }) returns { model: 'gpt-4o', source: 'global' } with no warning
• resolveModel({ globalDefaultModel: 'truly-unknown-model' }) still returns a warning

model-prompt.ts

• getModelChoices() returns at least 7 choices (3 Anthropic + 2 OpenAI minimum + custom)
• At least one choice has value starting with gpt-
• At least one choice has value starting with gemini-
• A __custom__ escape-hatch choice exists
• promptModelSelection() handles __custom__ by triggering a free-text input prompt
• DEFAULT_MODEL_CHOICE still defaults to claude-sonnet-4-20250514

---

Constraints

• No breaking changes: DEFAULT_MODEL stays as Claude Sonnet 4. Existing Claude users are unaffected.
• No new dependencies: Use existing @inquirer/prompts (select + input) for the custom model prompt.
• TDD: Write failing tests first (RED), then implement (GREEN), then refactor.
• Test coverage: Maintain ≥ 90% coverage on all changed files.

---

Context

This is a pure model-system change. It does not affect:

• The MCP protocol handlers
• Agent definitions in .ai-rules/agents/
• The config schema (users already set ai.defaultModel freely; this just removes false warnings and improves CLI UX)
• crush.json / opencode.json (personal dev config files, out of scope)

The resolveModel() function already supports unknown models gracefully (passes them through with a warning). This issue simply expands the "known" list and improves the CLI selection experience.