Bug_202_EVALUATE: Test Case ORCH-QA-001 — Whitespace-only task_summary stored in _workflow_context

[steadhac]

Component: finbot/agents/orchestrator.py → OrchestratorAgent._capture_agent_context (line 421)

Root cause:

# orchestrator.py line 421
if summary:
    self._workflow_context.append((agent_label, summary))

" " (whitespace-only string) is truthy in Python. if summary: evaluates to True when
summary=" ", so a meaningless blank summary is appended to _workflow_context and
propagated to downstream agents via _enrich_with_prior_context.

Steps to reproduce:

1. Create an OrchestratorAgent.
2. Call _capture_agent_context("invoice_agent", {"task_summary": " "}).
3. Inspect agent._workflow_context.

Expected: _workflow_context == [] — whitespace-only summary is not stored.
Actual: _workflow_context == [('invoice_agent', ' ')]

How to execute:

pytest tests/unit/agents/test_orchestrator.py::TestQAFindings::test_orch_qa_001_whitespace_only_summary_should_not_be_captured -v

Proposed fix:

# Before (buggy):
if summary:

# After (correct):
if summary and summary.strip():

Impact: A whitespace-only summary from any upstream agent is silently stored and injected
into the task description of every subsequent downstream agent. This pollutes the LLM context
with empty noise, wastes tokens, and — in a prompt injection scenario — could be used to slip
a blank payload through the if summary: guard undetected (the guard is bypassed by a
non-empty but content-free string).

Acceptance criteria:

• test_orch_qa_001_whitespace_only_summary_should_not_be_captured passes
• _capture_agent_context with task_summary=" " leaves _workflow_context empty
• All other _capture_agent_context and _enrich_with_prior_context tests continue to pass