Pi provider sends developer role to strict OpenAI-compatible upstreams

OpenClaw Version | OpenClaw 版本

N/A — Pi 0.84.1

Plugin Version | 插件版本

feat/server_team @ 0b7bc09; pi-plugin 0.1.0

Operating System | 操作系统

Windows 11

System Specification | 系统配置

No response

Describe the bug | 问题描述

The first-class Pi provider registers a reasoning OpenAI Chat Completions model without setting compat.supportsDeveloperRole.

Because the configured base URL is the local MemoryProxy URL rather than the actual upstream URL, Pi does not detect strict upstreams such as DeepSeek. It sends the system prompt with role: "developer".

Strict OpenAI-compatible upstreams reject that role with HTTP 400, so Pi gets no model response and the downstream tool, memory capture, and skill-mining flow cannot proceed.

To Reproduce | 复现步骤

1. Start MemoryProxy with a strict OpenAI-compatible upstream, such as DeepSeek.
2. Configure Pi 0.84.1 to load MemoryCore/pi-plugin and use the tdai provider.
3. Send a normal prompt, for example reply OK.
4. Inspect the forwarded OpenAI Chat Completions request or upstream response.

The request contains a developer system-prompt message and the strict upstream returns HTTP 400.

Expected behavior | 预期行为

The Pi TDAI provider should explicitly set compat.supportsDeveloperRole: false, causing Pi to send the system prompt as role: "system" for proxy-routed requests.

Error Logs / Screenshots | 报错日志/截图

No response

Additional context | 补充信息

This is independent of #1421 skill sync. It was found during real Pi/Proxy validation and belongs to the existing Pi provider integration.