Problem

In both api/core/agent/fc_agent_runner.py and api/core/agent/cot_agent_runner.py, the "update prompt tool" loop calls tool_instances[prompt_tool.name] without first checking if prompt_tool.name is a key in tool_instances.

# fc_agent_runner.py ~line 302
for prompt_tool in prompt_messages_tools:
    self.update_prompt_message_tool(tool_instances[prompt_tool.name], prompt_tool)

This raises an unhandled KeyError when a tool present in prompt_messages_tools is not found in tool_instances. While _init_prompt_tools() normally keeps them in sync, tool name collisions or future refactors could break this assumption. The safe pattern—used elsewhere in the codebase (e.g. _handle_invoke_action)—is to guard with a .get() check and skip missing tools.

Steps to Reproduce

1. Configure an agent app with a mix of API tools and dataset retrieval tools.
2. Trigger a scenario where the tool name in prompt_messages_tools diverges from the corresponding key in tool_instances (e.g. name mismatch introduced by a plugin update).
3. Observe an unhandled KeyError crashing the agent run.

Expected

When a prompt tool name is not found in tool_instances, the update should be skipped gracefully instead of raising KeyError.

Fix

Replace the bare dict lookup with a .get() guard:

for prompt_tool in prompt_messages_tools:
    tool_instance = tool_instances.get(prompt_tool.name)
    if tool_instance:
        self.update_prompt_message_tool(tool_instance, prompt_tool)

Apply the same fix in both fc_agent_runner.py and cot_agent_runner.py.