feat: add OpenAI/Anthropic tools support with tool emulation

- Parse tools/tool_choice from OpenAI and Anthropic requests
- Inject tool definitions into system prompt via toolemulation
- Parse action blocks (```json action) from model responses
- Retry logic for forced tool_choice (any/required)
- Return proper tool_calls / tool_use in responses
- Support streaming tools via collect-and-replay pattern
- Add tool history projection (assistant tool_calls + tool results)
- Model ID normalization: use official names (Qwen3.6-Plus, etc.)
- Fix resolveSessionMode to use Fresh mode when tools present

README.md

@@ -14,6 +14,8 @@ Current scope:
 - one request at a time
 - supports Windows named-pipe transport and local websocket transport
 - directly uses Lingma IPC, not DOM/CDP
+- OpenAI-compatible `tools` / `tool_choice` support (tool emulation via prompt engineering)
+- Anthropic-compatible `tools` / `tool_choice` support
 
 ## Run
 
@@ -69,6 +71,41 @@ Recommended layout:
 }
 ```
 
+## macOS / Linux
+
+This project also works on macOS and Linux via **WebSocket transport**. The Windows named-pipe transport is automatically skipped on non-Windows platforms.
+
+### Run on macOS
+
+```bash
+cd ~/OpenSources/lingma-ipc-proxy
+go run ./cmd/lingma-ipc-proxy --transport websocket --port 8095
+
+# Or use auto-detect (will discover websocket port from Lingma's shared client cache)
+go run ./cmd/lingma-ipc-proxy --port 8095
+```
+
+### Build on macOS / Linux
+
+```bash
+cd ~/OpenSources/lingma-ipc-proxy
+go build -o ./dist/lingma-ipc-proxy ./cmd/lingma-ipc-proxy
+```
+
+### macOS Config Example
+
+```json
+{
+  "host": "127.0.0.1",
+  "port": 8095,
+  "transport": "websocket",
+  "mode": "agent",
+  "shell_type": "zsh",
+  "session_mode": "auto",
+  "timeout": 120
+}
+```
+
 ## Build
 
 Build a Windows executable: