agent-lab/docs/plans/todo-8-tool-pre-message-guard.md

Todo 8 Tool Pre-Message Guard Plan

Status: done

Result

Implemented by subagent and locally verified by the main session.

Commands:

uv run pytest tests/test_websocket_api.py
uv run pytest

Results:

• focused WebSocket/API tests: 10 passed, 1 warning in 0.22s
• full suite: 16 passed, 1 warning in 0.22s

Goal

Prevent the debug UI and provider serializer from generating invalid Chat Completions tool messages without tool_call_id.

Problem

The pre-message editor exposes tool as a role, but the UI does not collect tool_call_id. A manual tool pre-message can therefore reach the provider as {role: "tool", content: "..."}, which strict compatible providers reject.

Scope

This todo covers:

• Removing tool from the static pre-message role selector for MVP.
• Adding provider-client validation that raises a clear error if a tool message lacks tool_call_id.
• Tests for the validation and UI markup.

This todo does not cover:

• Adding full manual tool-call transcript editing.
• New EventAgent tools.
• Live LLM smoke.

Files

• Modify: src/agent_lab/presentation/static/index.html
• Modify: src/agent_lab/infrastructure/chat_client.py
• Modify/add tests, preferably tests/test_websocket_api.py.
• Do not modify: docs/plans/todos.md.

Steps

1. Add a failing test that OpenAICompatibleChatClient rejects a ChatMessage(role="tool", content="...") without tool_call_id.
2. Add a failing test or assertion that the static pre-message role selector does not offer tool.
3. Implement the minimal fix:
   • remove <option value="tool">tool</option> from the pre-message template.
   • in _serialize_message, raise ValueError("tool messages require tool_call_id") when role is tool and no tool_call_id is present.

4. Run:

   uv run pytest tests/test_websocket_api.py
   uv run pytest
   

Verification

Expected result:

• Invalid manual tool messages fail before network I/O.
• UI no longer exposes invalid manual tool pre-message creation.
• Full suite passes.