[ervareza]

Problem

When validating custom models in Claude Code CLI (e.g. /model cbai/glm-5.2), Claude Code sends a non-streaming (stream: false) POST request to /v1/messages.

Upstream OpenAI-compatible providers return standard OpenAI JSON ({"choices": [...], "usage": {"prompt_tokens": 751}}). 9router's format resolution defaulted to openai because:

1. detectFormatByEndpoint was attempting new URL(request.url) on relative endpoints, throwing an unhandled TypeError: Invalid URL.
2. detectFormat(body) did not inspect the endpoint or anthropic-version header, falling back to openai.

As a result, 9router returned raw OpenAI format JSON without converting it to Anthropic format ({"type": "message", "usage": {"input_tokens": 751}}), causing Claude Code CLI to crash with:
Unable to validate model: undefined is not an object (evaluating 'z.usage.input_tokens')

Fix

• open-sse/translator/formats.js: Update detectFormatByEndpoint to match /messages endpoints and safely handle null/relative paths.
• src/sse/handlers/chat.js: Extract endpoint safely from clientRawRequest?.endpoint before trying request?.url to avoid Invalid URL parsing errors.
• open-sse/handlers/chatCore.js: Set fallback sourceFormat to FORMATS.CLAUDE whenever the endpoint contains /messages.
• open-sse/services/provider.js: Add support for anthropic-version header in format detector.

Tests

• Verified via non-streaming POST http://127.0.0.1:20128/v1/messages with cbai/glm-5.2:
  • Returns HTTP 200 OK with type: "message" and usage: { input_tokens, output_tokens }.
• Verified model validation in Claude Code CLI passes cleanly.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.