Filter the $schema field in tool parameters of kosong genai provide…

[pvzheroes125]

Filter the $schema field in tool parameters of kosong genai provider so that kimi-cli can work properly when using gemini models with third-party mcp. (Tool parameters is a BaseModel with extra=forbid in genai sdk while TypedDict in anthropic and openai sdk)

Related Issue

Resolve #(issue_number)

Description

Checklist

• I have read the CONTRIBUTING document.
• I have linked the related issue, if any.
• I have added tests that prove my fix is effective or that my feature works.
• I have run make gen-changelog to update the changelog.
• I have run make gen-docs to update the user documentation.

---

[devin-ai-integration]

Devin Review found 1 potential issue.

View issue and 1 additional flag in Devin Review.

[devin-ai-integration]

🔴 Mutation of shared tool.parameters dict removes $schema field permanently

The tool_to_google_genai function directly mutates the tool.parameters dictionary by deleting the $schema key, which permanently modifies the original Tool object.

Click to expand

How the bug is triggered

The Tool object is a Pydantic BaseModel that is typically reused across multiple API calls. When tool_to_google_genai is called:

if '$schema' in tool.parameters:
    del tool.parameters['$schema']

This directly modifies the original tool.parameters dict. Since the same Tool instance is returned by toolset.tools (see kosong/tooling/simple.py:89-90), subsequent uses of this tool will have the $schema field missing.

Actual vs Expected behavior

Actual: The $schema field is permanently removed from the original tool object after the first call to tool_to_google_genai.

Expected: The function should create a copy of the parameters dict before modifying it, leaving the original tool object unchanged.

Impact

1. If the same tool is used with multiple providers (e.g., OpenAI and Google GenAI), the $schema field will be missing for subsequent provider calls.
2. If the tool's parameters dict is used elsewhere in the application (e.g., for validation, logging, or serialization), it will be missing the $schema field.
3. This violates the principle of least surprise - a conversion function should not have side effects on its input.

Recommendation: Create a shallow copy of the parameters dict before modifying it:

parameters = tool.parameters.copy()
if '$schema' in parameters:
    del parameters['$schema']
return Tool(
    function_declarations=[
        FunctionDeclaration(
            name=tool.name,
            description=tool.description,
            parameters=parameters,
        )
    ]
)

---

Was this helpful? React with 👍 or 👎 to provide feedback.