Handle non-JSON (HTML) responses from OpenAI-compatible models endpoint

[Copilot]

When an OpenAI-compatible server returns HTML instead of JSON for the /models endpoint (e.g. a misconfigured proxy or gateway), goose crashes with an opaque parse error.

Changes

• handle_response_openai_compat: Read body as text before parsing JSON. When parsing fails and the body starts with <, include a diagnostic hint in the error: "response appears to be HTML — check that the API endpoint URL is correct".

• OpenAiCompatibleProvider::fetch_supported_models: Catch RequestFailed errors (non-JSON/HTML 200s, missing endpoints) and return Ok(vec![]) with a warn! log instead of propagating. Auth, rate-limit, and server errors still propagate normally.

• Unit test: wiremock-backed test asserting an HTML 200 OK on GET /models yields Ok(vec![]).

Summary

Type of Change

• Feature
• Bug fix
• Refactor / Code quality
• Performance improvement
• Documentation
• Tests
• Security fix
• Build / Release
• Other (specify below)

AI Assistance

• This PR was created or reviewed with AI assistance

Testing

New #[tokio::test] in openai_compatible.rs using wiremock: spins up a mock server returning HTML for GET /models, asserts fetch_supported_models() returns Ok(vec![]). All existing openai_compatible tests continue to pass.

Related Issues

Screenshots/Demos (for UX changes)

Before:

After:

Original prompt

sometimes an openai server responds with HTML for the root endpoint. right now goose dies when that is the case , update goose to be robust to receiving non-json responses from the root endpoint. add a unit test for it too

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.