refactor(models): Refine MessageAgentThought SQLAlchemy typing

[tomerqodo]

Benchmark PR from qodo-benchmark#425

---

Note

Modernizes agent thought persistence and history reconstruction.

• Refactors MessageAgentThought to extend TypeBase with Mapped[...] typing, insert_default/default_factory ID generation, Decimal pricing fields, explicit defaults, and reordered required fields (created_by_role/created_by)
• Updates BaseAgentRunner to:
  • Create thoughts using Decimal prices and CreatorUserRole enum; add tool_process_data default
  • Safely append thought text; handle missing/invalid JSON for tool_input/observation when rebuilding history
  • Build tool call lists from tool names more defensively and handle empty cases
• Adds AgentThoughtValidation (Pydantic) model scaffold
• Tests updated to rely on auto-generated MessageAgentThought.id and align with new model behavior

^{Written by Cursor Bugbot for commit 8821d28. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[cursor]

Tool response replaced with tool input in history

High Severity

The ToolPromptMessage content now uses tool_inputs.get(tool, ...) instead of tool_responses.get(tool, ...). This means tool outputs/responses are replaced with tool inputs when reconstructing agent conversation history. The LLM will receive incorrect context where tool call responses show what was sent to the tool rather than what was returned by the tool.

 

[cursor]

Answer price unit values are swapped

Medium Severity

The answer_unit_price and answer_price_unit values appear swapped. The message fields use message_unit_price=Decimal(0) and message_price_unit=Decimal("0.001"), but the answer fields use the opposite pattern: answer_unit_price=Decimal("0.001") and answer_price_unit=Decimal(0). The model defaults both *_price_unit fields to Decimal("0.001").