Pure instruction skill. LLM structured thinking + read_draft, search_local_references, mcp_cgu_deep_think, mcp_cgu_spark_collision。
Triggers: debate, pro/con, devil's advocate, challenge assumptions, compare viewpoints
Framework 1: Academic Debate (two-sided debate)
Structured pro and con arguments for a research issue. Bylevel of evidenceranked: SR/Meta > RCT > Cohort > Case-Control > Cross-Sectional > Case Series > Expert Opinion
Output: Position A (supporting) → Position B (opposing) → Methodological Considerations (bias) → Synthesis (agreement / disagreement / clinical bottom line)
Framework 2: Devil's Advocate
Systematically challenge a research claim. 5 dimensions:
- Methodological — biases corresponding to study type (see below)
- Statistical — multiple comparisons, effect size, CI width, missing data, power
- Generalizability — population, setting, timeframe, intervention fidelity
- Alternative Explanations — confounding, reverse causation, temporal trends, Hawthorne
- Likely Reviewer Questions — based on study type
Output: Strengthening Recommendations (address counter-arguments, add sensitivity analysis, cite SR, frame conservatively)
Framework 3: Viewpoint Comparison
Systematic comparison of multiple theories/methods. Default comparison dimensions: Evidence Base, Theoretical Foundation, Clinical Applicability, Patient Safety, Cost-Effectiveness, Current Guidelines, Limitations
Study-Type-Specific Biases
| Type | Key Biases |
|---|---|
| RCT | Selection, performance, detection, attrition, reporting |
| Cohort | Selection, confounding, information, loss to follow-up, healthy worker |
| Case-Control | Recall, selection (controls), confounding, misclassification, temporal ambiguity |
| Cross-Sectional | No temporality, prevalence-incidence, non-response, information |
| Retrospective | Information (records), survivorship, confounding, missing data |
Auto-detect：randomized/RCT=RCT, cohort/prospective=Cohort, case-control/odds=Case-Control, cross-sectional/prevalence=Cross-Sectional, retrospective/chart=Retrospective
Principles of use
- Evidence-based, cite existing literature
- Present both sides in a balanced way
- Bias analysis corresponds to the correct study type
- Finally provide clinical recommendations (in service of the Discussion)