ponytail-audit: no pass to flag AI-shaped test brittleness (fixture-name branches in prod, snapshot/mock-only assertions, 1:1 method-mirror tests) #683

ponytail's test stance is minimal and construction-time: "non-trivial logic leaves one runnable check." But ponytail-audit, scanning existing code, has no pass to flag AI-generated test slop, which is a distinct and common failure mode:

• tests that mirror implementation methods 1:1 (break on any refactor; they assert structure, not behavior)
• snapshot-only or mock-call-only assertions (they pin the AI's own output, not observable behavior)
• worst: fixture-name / id / timestamp branches leaking into production code to make a test pass

These pass CI and expose no deletable abstraction, so the ladder skips them, but they're maintainability slop, and the fixture-branch-in-prod case is also a correctness smell (production behavior special-cased for test data).

Distinct from #602, which is about ponytail constraining the agent while writing tests; this is the audit skill detecting brittleness in tests that already exist.

Suggested: one ## Hunt bullet: fixture-name/id/timestamp branches in production code (encode the real invariant instead); snapshot-only / mock-call-only assertions (prefer observable behavior); tests that mirror methods 1:1 rather than behaviors.

Source: Google Testing Blog, "Test Behaviors, Not Methods."

Happy to open the PR.