refactor(models): Refine MessageAgentThought SQLAlchemy typing

[tomerqodo]

Benchmark PR from agentic-review-benchmarks#11

---

Note

Modernizes agent thought persistence and improves history reconstruction.

• Refactors MessageAgentThought to extend TypeBase with typed Mapped fields, insert_default/default_factory IDs, nullable defaults, and Decimal pricing with sensible server defaults
• Introduces created_by_role/created_by typing on the model and uses CreatorUserRole enum in runner
• Base agent runner: safer thought concatenation, uses Decimal prices, null-safe/defensive JSON parsing for tool_input/observation, and clearer tool name handling; constructs tool_calls/responses even with empty or malformed payloads
• Adds lightweight AgentThoughtValidation model (pydantic) for pre-persist validation scaffold
• Tests updated to rely on auto-generated IDs and align with new model behavior

^{Written by Cursor Bugbot for commit 5209691. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 3 potential issues.

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

[cursor]

Tool response data replaced with tool input data

High Severity

The ToolPromptMessage content incorrectly uses tool_inputs instead of tool_responses. The code parses tool_responses from agent_thought.observation (lines 478-485) but then mistakenly uses tool_inputs.get(tool, ...) when constructing the response message. This causes agent history to contain tool inputs where tool outputs/responses should be, corrupting the conversation context for subsequent agent interactions.

 

[cursor]

Answer price unit values are swapped

Medium Severity

The answer_unit_price and answer_price_unit values appear to be swapped. The pattern for message_* fields shows message_unit_price=Decimal(0) and message_price_unit=Decimal("0.001"), but for answer_* it's reversed: answer_unit_price=Decimal("0.001") and answer_price_unit=Decimal(0). This is inconsistent with the model's default of Decimal("0.001") for answer_price_unit and could cause incorrect price calculations.

 

[cursor]

SQL text object used as Python dataclass default

Medium Severity

Changing MessageAgentThought from Base to TypeBase (which uses MappedAsDataclass) makes default=sa.text("'{}'") problematic. In dataclass mode, the default parameter provides the Python default value, so tool_labels_str and tool_meta_str would default to a TextClause object instead of the string "{}". The correct pattern (as used in other TypeBase classes like Account) is default="{}", server_default=sa.text("'{}'").