refactor(models): Refine MessageAgentThought SQLAlchemy typing

[tomerqodo]

Benchmark PR from qodo-benchmark#425

[sentry]

Bug: The agent history incorrectly uses tool_inputs instead of tool_responses for ToolPromptMessage content, providing the agent with tool inputs instead of outputs.
_{Severity: HIGH}

Suggested Fix

In base_agent_runner.py on line 502, change content=str(tool_inputs.get(tool, agent_thought.observation)) to use the tool_responses variable, which holds the actual tool output. The line should be content=str(tool_responses.get(tool, agent_thought.observation)).

Prompt for AI Agent

Review the code at the location below. A potential bug has been identified by an AI
agent.
Verify if this is a real issue. If it is, propose a fix; if not, explain why it's not
valid.

Location: api/core/agent/base_agent_runner.py#L502

Potential issue: In the `organize_agent_history` method, the content for
`ToolPromptMessage` is sourced from `tool_inputs` instead of the correct
`tool_responses` variable. This causes the agent's conversation history to record the
input parameters sent to a tool, rather than the actual output returned by the tool. As
a result, the language model will not have the necessary information from previous tool
executions to make correct decisions in subsequent steps, breaking the agent's reasoning
process in multi-turn interactions.

_{Did we get this right? 👍 / 👎 to inform future reviews.}

[sentry]

Bug: The ToolPromptMessage is incorrectly created with tool inputs from tool_inputs instead of tool outputs from tool_responses, sending the wrong information back to the agent.
_{Severity: HIGH}

Suggested Fix

In api/core/agent/base_agent_runner.py on line 502, change the content parameter for ToolPromptMessage to use the tool_responses dictionary. Replace str(tool_inputs.get(tool, agent_thought.observation)) with str(tool_responses.get(tool, agent_thought.observation)).

Prompt for AI Agent

Review the code at the location below. A potential bug has been identified by an AI
agent.
Verify if this is a real issue. If it is, propose a fix; if not, explain why it's not
valid.

Location: api/core/agent/base_agent_runner.py#L502

Potential issue: When creating the `ToolPromptMessage` for the agent's conversation
history, the code incorrectly uses `tool_inputs.get(...)` instead of the intended
`tool_responses.get(...)`. This causes the tool's input arguments to be sent back to the
agent as if they were the tool's output. As a result, the agent lacks the actual results
from the tool execution, which breaks its reasoning chain and prevents it from
functioning correctly.

_{Did we get this right? 👍 / 👎 to inform future reviews.}