adversarial-review

Use when choosing or changing the structure of a research experiment repository, based on the current experiment's actual needs.

Use when a research task crosses domains or its required evidence is unclear.

Use for multi-session or handover-driven work, or substantial claim-bearing work that needs recoverable state beyond existing run records.

Use before a semantic acceptance review to choose its scope and stop repeated verdicts without limiting evidence-driven result analysis.

Use when writing Korean research reports, result interpretations, experiment logs, or substantive reproducibility and handover documents.

Use for physical instrument/board acquisition, preflight, trigger integrity, and raw capture evidence.

Use to accept a research result, benchmark, or material implementation/report claim against its evidence.

Use to coordinate independent agent deliverables, concurrent local compute, or long-running jobs.

Use for comment/docstring audits or a suspected mismatch between an explanation and current code.

Use Sequential Thinking MCP for genuinely hard planning, unclear debugging, expensive experiment design, and claim acceptance.

Use for side-channel attack research, leakage detection, power/EM traces, classifier validation, key recovery, and countermeasure evaluation.

Use to design or assess AI/ML experiments, evaluation datasets, baselines, metrics, and reproducibility.