What problem does it solve? Existing tests only prove the agent pipeline runs (status codes, executed flags) but never prove the answer is right, so a hallucinated balance passes every test. This Skill closes that gap by asserting actual values against ground truth. ## Core Features & Use Cases - Value-level assertions: Assert routing params, tool payloads, and rendered narration against the seed/store or that turn's /api/mcp/tool payload instead of status codes. - Trap detection: Avoids false-green tests caused by the heuristic floor answering chips before the LLM, dual result shapes (banking vs vertical), and per-vertical money key differences like amountDue in healthcare. - Ground truth mapping: Locates each vertical's real data source (seed.json files, GET /api/accounts/my) and its own primaryTool contract in useCases.js. - Use Case: Before a demo, verify that the chip "pay my $300 bill" routes with params.amount === 300 and that the rendered reply matches the tool payload, in both heuristic and LLM-only modes. ## Quick Start Verify that the healthcare chips return real billing values from the seed data and that the LLM actually ran by checking the llm-proxy logs.
