skill: diagnosing-bugs
tasks:
  - prompt: "Diagnose this: the API returns 500 every Friday afternoon but works fine other days"
    grader:
      - the response invokes the diagnosing-bugs skill
      - the response asks for or constructs a feedback loop (failing test, repro script, captured trace)
      - the response does not jump to a fix before establishing the loop
  - prompt: "My tests have started failing intermittently after the v1.4 deploy"
    grader:
      - the response invokes the diagnosing-bugs skill
      - the response addresses determinism (seeded RNG, time pinning, isolated env)
  - prompt: "Write a haiku about autumn"
    grader:
      - the response does NOT invoke the diagnosing-bugs skill