OpenClaw not sending POST /v1/chat/completions after model switch

[igorrost1909-eng]

When using gemini-web2api as an OpenAI-compatible proxy with OpenClaw:

• Proxy works perfectly via curl: /v1/models, non-stream chat, streaming chat, streaming+tools → all 200 OK
• HTTP/1.1 + keep-alive headers added to server.py
• Auth disabled
• OpenClaw plugin extension created with modelCatalog + baseUrl
• OpenClaw correctly switches model to gemini-web2api/gemini-3.5-flash (verified via session_status)
• But when the agent tries to respond → "Agent couldn't generate a response"
• Dump/log proxy shows only GET /v1/models (200), NO POST /v1/chat/completions ever arrives

Hypothesis: OpenClaw registers the provider but doesn't route requests without a specific providerEndpoint structure. Possibly needs endpointClass or a specific plugin format.

Setup: Python 3.12, gemini-web2api latest commit, config with default cookie. OpenClaw 2026.6.11.