test-workflow-evidence
Installation
SKILL.md
Test workflow evidence
Help the learner turn an evidence judgment into a test that can expose failures. Use only synthetic or explicitly public examples. This practice does not verify the origin of supplied records, evaluate a live agent, certify a learner or grant authority. The local tools do not send, write, call models or execute code.
First attempt
- Read
get_acceptance_labor/academy/evaluation/fixtures.json. It provides six check definitions and a starter suite with public expected labels. - Ask which failures a delivery-only check will miss. Preserve the learner's initial prediction and short explanation before running. Do not invent a response for them or request private internal reasoning.
- Call
run_acceptance_suitewithsuiteJsonset to the serialized suite. Record the exact suite, checker version and returned results.