feat: refactor sgclaw around zeroclaw compat runtime

This commit is contained in:
zyl
2026-03-26 16:23:31 +08:00
parent bca5b75801
commit ff0771a83f
1059 changed files with 409460 additions and 23 deletions

View File

@@ -0,0 +1,66 @@
{
"model_name": "test-multi-tool-chain",
"turns": [
{
"user_input": "Echo three messages in sequence",
"steps": [
{
"response": {
"type": "tool_calls",
"tool_calls": [
{
"id": "call_1",
"name": "echo",
"arguments": {"message": "first"}
}
],
"input_tokens": 30,
"output_tokens": 15
}
},
{
"response": {
"type": "tool_calls",
"tool_calls": [
{
"id": "call_2",
"name": "echo",
"arguments": {"message": "second"}
}
],
"input_tokens": 60,
"output_tokens": 15
}
},
{
"response": {
"type": "tool_calls",
"tool_calls": [
{
"id": "call_3",
"name": "echo",
"arguments": {"message": "third"}
}
],
"input_tokens": 90,
"output_tokens": 15
}
},
{
"response": {
"type": "text",
"content": "I echoed three messages: first, second, and third.",
"input_tokens": 120,
"output_tokens": 20
}
}
]
}
],
"expects": {
"response_contains": ["first", "second", "third"],
"tools_used": ["echo"],
"max_tool_calls": 3,
"all_tools_succeeded": true
}
}

View File

@@ -0,0 +1,38 @@
{
"model_name": "test-single-tool-echo",
"turns": [
{
"user_input": "Echo hello for me",
"steps": [
{
"response": {
"type": "tool_calls",
"tool_calls": [
{
"id": "call_1",
"name": "echo",
"arguments": {"message": "hello"}
}
],
"input_tokens": 30,
"output_tokens": 15
}
},
{
"response": {
"type": "text",
"content": "The echo tool said: hello",
"input_tokens": 50,
"output_tokens": 10
}
}
]
}
],
"expects": {
"response_contains": ["hello"],
"tools_used": ["echo"],
"max_tool_calls": 1,
"all_tools_succeeded": true
}
}

View File

@@ -0,0 +1,24 @@
{
"model_name": "test-smoke-greeting",
"turns": [
{
"user_input": "Hello, how are you?",
"steps": [
{
"response": {
"type": "text",
"content": "Hello! I'm doing well, thank you for asking. How can I help you today?",
"input_tokens": 20,
"output_tokens": 15
}
}
]
}
],
"expects": {
"response_contains": ["Hello"],
"response_not_contains": ["error", "ERROR"],
"tools_used": [],
"max_tool_calls": 0
}
}