Bug_196_EVALUATE: CHAT-BOUNDARY-013 — Both required fields None crash agent before any task is dispatched

[steadhac]

Component: finbot/agents/chat.py → VendorChatAssistant._call_start_workflow (line 229)

Root cause:

Same as CHAT-BOUNDARY-012. When both description=None and vendor_id=None are passed,
description[:100] on line 229 crashes first. The crash masks the vendor_id=None gap
(TICKET 9) — neither field is validated before the exception is raised.

Steps to reproduce:

1. Call _call_start_workflow(None, vendor_id=None).

Expected: {"error": "description and vendor_id are required"}
Actual: TypeError: 'NoneType' object is not subscriptable

How to execute:

pytest tests/unit/agents/test_chat_assistant.py::TestBoundaryAndTypeValues::test_chat_boundary_013_both_required_fields_none_crashes -v

Proposed fix: Same unified guard as TICKET 10 — check both fields at the top of
the method before any other logic runs.

Impact: The most degenerate possible LLM tool call — all required fields absent —
produces an unhandled crash rather than a recoverable error. Resolving TICKETS 9, 10,
and 11 together with a single early-return guard covers all combinations.

Acceptance criteria:

• test_chat_boundary_013_both_required_fields_none_crashes updated to assert error JSON
• Single guard at method entry handles description is None OR vendor_id is None

[Jean-Regis-M]

@saikishu @steadhac The crash occurred because None values bypassed the pre‑dispatch validation layer. I added a strict type check at the workflow entry point, which now returns a structured error instead of raising a TypeError. This stabilizes the agent loop without altering the core logic.