fix: guard vendor_id/invoice_id=None in chat assistant tool callables (#418)

[Deez-Automations]

Fixes #418.

The bug

VendorChatAssistant and CoPilotAssistant's native tool callables forward vendor_id/invoice_id straight to the underlying data functions with no None check. If the LLM calls a tool without supplying the ID argument, it reaches the data layer as None.

Corrections made against the linked issue

• Scope is 12 call sites, not 5. The issue lists 5 methods on one class. In practice the same 5 methods are duplicated byte-for-byte across both VendorChatAssistant (vendor-portal-facing, POST /chat) and CoPilotAssistant (admin-portal-facing, POST /copilot/chat) — 10 sites total. Auditing the rest of _build_native_callables() for the same pattern turned up two more CoPilotAssistant-only methods with the identical gap: _call_get_vendor_compliance_docs and _call_get_vendor_activity_report. 12 sites fixed total.
• Exception type correction. The issue says a TypeError (or a raw SQLAlchemy error) propagates. Traced the actual behavior instead of trusting that: SQLAlchemy translates Column == None into a valid IS NULL clause, so e.g. VendorRepository.get_vendor(None) just returns no matching row (primary keys are never NULL) rather than raising anything at the DB layer. It's the application code's own if not X: raise ValueError(...) checks that actually fire. Every affected path raises ValueError, never TypeError.
• Severity correction. Every one of these methods is called exclusively through _execute_tool, which already wraps the call in a broad try/except Exception and converts any exception into a clean JSON error string before it reaches the LLM. So a None ID here was never an uncaught crash — the real (still worth fixing) bug is a misleading error message (e.g. "Vendor not found" when the actual problem is "you forgot to pass vendor_id"), which risks the LLM reasoning incorrectly, like concluding a real vendor doesn't exist.

Fix

Added an explicit is None guard at the top of all 12 methods, returning {"error": "<field> is required"} before the call reaches the data layer, and widened the type hints to int | None to reflect reality.

Tests

13 new tests in tests/unit/agents/test_chat_assistant_none_guards.py: one per guarded method (5 on VendorChatAssistant, 7 on CoPilotAssistant), plus one integration test through _execute_tool documenting the severity-correction point above. Confirmed RED before the fix, GREEN after.

Test plan

• 13 new unit tests pass
• Broader regression (tests/unit/agents/, tests/unit/ctf/) run — 3 pre-existing failures in test_specialized_agents.py confirmed unrelated (different agent classes, reproduced on a clean stash of this branch's change, i.e. present on main without this fix)
• Local merge-tested solo against fresh origin/main — clean
• Local merge-tested pairwise against Fix: orchestrator confirms payment to vendor even when it fails #574, Fix: DNS-based SSRF bypass in guardrail webhook URL validation #575, Fix: guardrail payload corruption before signing + missing after_tool for complete_task #576, Fix: Dark Lab tool override silently drops parameters/inputSchema #577, Fix: profile share-card cache key omits fields that affect the render #578 — all clean

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.