Bug: ValueError crash when LLM outputs "args" instead of "tool_args"

[wgnrai]

Bug Report

Problem

In agent.py, process_tools() calls validate_tool_request(tool_request) without normalizing the dictionary first. Some LLMs output "args" instead of "tool_args".

This causes a crash:

ValueError: Tool request must have a tool_args (type dictionary) field

Location

agent.py line ~877-878:

if tool_request is not None:
    await self.validate_tool_request(tool_request)

Reproduction

1. Use an LLM that outputs "args" key instead of "tool_args"
2. Tool request JSON: {"tool_name": "some_tool", "args": {...}}
3. validate_tool_request() raises ValueError

Proposed Fix

Add normalization before validation:

if tool_request is not None:
    # Normalize: map "args" -> "tool_args", default to empty dict
    if "tool_args" not in tool_request:
        tool_request["tool_args"] = tool_request.get("args", {})
    if not isinstance(tool_request["tool_args"], dict):
        tool_request["tool_args"] = {}
    await self.validate_tool_request(tool_request)

Consistent with downstream handling at line ~882:

tool_args = tool_request.get("tool_args", tool_request.get("args", {}))

Impact

• Prevents crashes with LLMs using non-standard key naming
• No behavior change for correctly formatted requests
• Defensive programming for edge cases

PR ready with this fix.