#17263 (comment) demonstrates using the LLM to review the prompt for contradictory language.
Contradictions are notable because they lead to inconsistent or not working prompted behavior an are often a sign of deficiencies in the prompt.
This issue tracks adding a simple integration test that reads the prompt, enumerates all ambiguities or contradictions, and checks an allow list for them. Any that aren't present in the allow list, as judged by the LLM, should cause the agent to return a failure message and fail the test.