OpenCode Go: Muse/Grok/GPT-5.6-Luna models fail with 500 -- requires /responses and /messages endpoints, not just /chat/completions #1224

Problem

OpenCode Go is a multi-protocol provider. Different models require different API endpoints and formats:

jcode currently routes all OpenAI-compatible providers through /chat/completions (hardcoded in openrouter_sse_stream.rs line 180). This means Muse, Grok, and GPT-5.6-Luna models return HTTP 500, and MiniMax/Qwen models also fail.

Reproduction

1. Configure OpenCode Go provider with a valid API key
2. Select muse-spark-1.3-contributor model
3. Send any message
4. Observe HTTP 500 errors in logs:

API stream attempt 1/8 over HTTPS transport (model: muse-spark-1.3-contributor, endpoint: https://opencode.ai/zen/go/v1, auth: OPENCODE_GO_API_KEY)
status: 500 Internal Server Error
response: {"type":"error","error":{"type":"error","message":"Internal server error"}}

Direct curl confirmation:

• muse-spark-1.3-contributor via /chat/completions → HTTP 500
• muse-spark-1.3-contributor via /responses → HTTP 200
• kimi-k2.7-code via /chat/completions → HTTP 200

Root Cause

In crates/jcode-provider-openrouter-runtime/src/openrouter_sse_stream.rs:

let url = format!("{}/chat/completions", api_base);

This hardcodes /chat/completions for all OpenAI-compatible providers. OpenCode Go requires per-model endpoint routing.

Expected Behavior

jcode should route OpenCode Go models to their correct endpoints:

• Muse/Grok/GPT-5.6-Luna → /responses (OpenAI Responses API format)
• MiniMax/Qwen → /messages (Anthropic Messages API format)
• GLM/Kimi/DeepSeek/MiMo/Hy → /chat/completions (current behavior)

Related Issues

• Goose #11980 -- same issue in Goose
• OpenCode #43379 -- Muse streaming issues on /chat/completions

Workaround

Use models that work on /chat/completions: kimi-k2.7-code, glm-5.3, deepseek-v4-pro, etc.

Environment

• jcode v0.84.0 (57d5878)
• macOS aarch64
• OpenCode Go provider with valid API key