[Bug]: Issues encountered when running some model examples

[kamuiiiii]

Describe the bug

Provider	Model	Example	Error
openai ✅	o3-mini	workflow-orchestrator	error: 400 Unsupported parameter: 'parallel_tool_calls' is not supported with this model.
openai ➖	o1-mini	mcp-github	error: 400 Unsupported value: 'messages[0].role' does not support 'system' with this model.
anthropic	claude-3-7-sonnet-latest	mcp-sqlite	error: {"type":"error","error":{"details":null,"type":"overloaded_error","message":"Overloaded"} }
anthropic ➖	claude-3-5-haiku-latest	workflow-orchestrator	error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"prompt is too long: 242843 tokens > 200000 maximum"}}
xai ➖	grok-2-latest	workflow-orchestrator	error: Max iterations reached: 3
openai ➖	gpt-4o-mini	workflow-orchestrator	error: Max iterations reached: 3
gemini ➖	gemini-2.0-flash	workflow-orchestrator	429 status code (no body) - “You've exceeded the rate limit.”
open router ✅	openai/gpt-4o	workflow-router	Router toolChoice requires exactly one tool to be executed (When toolChoice is set to router, an error is thrown if multiple tools are selected.)
openai ➖	o4-mini	workflow-code-execution	call agent js-sandbox failed with error: ReferenceError: module is not defined
gemini ➖	gemini-1.5-flash	mcp-github	FAILED: MCP error -32603: Authentication Failed: Requires authentication
ollama ➖	llama3.2	workflow-router	FAILED: Tool not found: main
ollama ➖	llama3.2	chat-bot	Connect error
openai ✅	o3-mini	workflow-reflection	error: Agent AIAgent output check arguments error: code: Required
openai ✅	o3-mini / o4-mini	chat-bot	FAILED: 400 Unsupported parameter: 'temperature' is not supported with this model.

Row 1

Row 2

Row 3

Row 4

Row 7

Row 8

Row 9

Row 10

Row 11

Row 13

Reproduction Steps

1. cd examples/...
2. CI=true MODEL_PROVIDER=... MODEL_NAME=... pnpm run start

Observed Behavior

An error is thrown during execution.

Expected Behavior

Return the expected results correctly.

Possible Solution

No response

Additional Information/Context

No response

[wangshijun]

todo list to make aigne framework more stable: @li-yechao

• Support auto retry when overload, with max retry and backoff
• Support early return for models that does not support text or image, aigne should include basic model metadata
• Support token length check before call model

[kamuiiiii]

Max retry and backoff are supported by default in both the OpenAI and Anthropic SDKs, with a default value of 2.
https://github.com/openai/openai-node?tab=readme-ov-file#retries
https://github.com/anthropics/anthropic-sdk-typescript?tab=readme-ov-file#retries