refactor(providers): extract ProviderDef trait and OpenAiCompatibleProvider

[codefromthecrypt]

Summary

Separate provider construction from runtime behavior to reduce boilerplate and prepare for ACP-backed providers.

ProviderDef: new trait with metadata() and from_env(), split from Provider. Simplifies registration from a closure-based call to a single type parameter:

-        registry.register::<XaiProvider, _>(|m| Box::pin(XaiProvider::from_env(m)), false);
+        registry.register::<XaiProvider>(false);

The name follows Rust conventions (FromStr, Default) rather than OOP "factory" terminology. The implementing structs are zero-sized types used purely as type-level descriptors, with an associated type that ensures compile-time verification that the constructed provider implements Provider.

OpenAiCompatibleProvider: reusable struct that implements Provider for any OpenAI-compatible API. Providers like xAI and Azure go from ~200 lines (full Provider impl with duplicated HTTP/streaming logic) to ~70 lines (just metadata + client setup). Applied initially to "xai" and "azure" to prove the concept; others can follow.

This is vital for ACP-backed providers, as there could be over a dozen and differ only on aspects such as permission and mode mappings.

Type of Change

• Refactor / Code quality

AI Assistance

• This PR was created or reviewed with AI assistance

Testing

Ran the normal tests.

Related Issues

Last major refactor needed to reduce the huge diff in ACP provider PR #6605

[Copilot]

Pull request overview

This PR refactors the provider system by separating provider construction (ProviderFactory trait) from runtime behavior (Provider trait), reducing boilerplate for OpenAI-compatible providers.

Changes:

• Introduces ProviderFactory trait with metadata() and from_env() methods, previously part of Provider
• Adds OpenAiCompatibleProvider implementation to consolidate shared OpenAI-compatible logic
• Simplifies provider registration from registry.register::<XaiProvider, _>(|m| Box::pin(XaiProvider::from_env(m)), false) to registry.register::<XaiProvider>(false)

Reviewed changes

Copilot reviewed 37 out of 37 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
crates/goose/src/providers/base.rs	Adds ProviderFactory trait, removes metadata() from Provider trait
crates/goose/src/providers/openai_compatible.rs	New file containing shared OpenAI-compatible provider implementation and utility functions moved from utils.rs
crates/goose/src/providers/xai.rs	Refactored to use OpenAiCompatibleProvider, significantly reducing boilerplate
crates/goose/src/providers/azure.rs	Refactored to use OpenAiCompatibleProvider, removing duplicate implementation
crates/goose/src/providers/provider_registry.rs	Simplified registration using ProviderFactory trait
crates/goose/src/providers/factory.rs	Updated all provider registrations to use simplified syntax
crates/goose/src/providers/utils.rs	Removed OpenAI-compat utility functions (moved to openai_compatible.rs)
crates/goose/src/providers/openai.rs	Updated to use get_secrets() for better secret handling, implements ProviderFactory
crates/goose/src/providers/*.rs	All providers updated to implement ProviderFactory separately from Provider
crates/goose/tests/*.rs	Test mocks updated to implement ProviderFactory, removed metadata() from mock Provider impls
crates/goose-mcp/.gooseignore	Standard goose ignore file added for MCP crate
crates/goose-acp/tests/fixtures/mod.rs	Added recursion limit directive for consistency
clippy-baselines/too_many_lines.txt	Added build_canonical_models to allowed long functions

[Copilot]

Pull request overview

Copilot reviewed 36 out of 36 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 36 out of 36 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 37 out of 37 changed files in this pull request and generated no new comments.

[michaelneale]

I think this is ok - main concern was azure shift to streaming and missing methods for loading models @codefromthecrypt