u9401066 稿件審查與報告指引技能
LLM 語意理解 > regex 關鍵字匹配。資料用 read_draft MCP tool，審查邏輯由 LLM + domain knowledge。
觸發：審查草稿、reviewer、CONSORT、STROBE、PRISMA、CARE、投稿前檢查
搭配 MCP Tools
| 工具 | 用途 |
|---|---|
read_draft |
必用 — 讀草稿 |
list_drafts |
列出可審查的草稿 |
count_words |
字數統計 |
check_formatting |
期刊格式量化檢查 + 跨文件一致性 |
Framework 1: Peer Review
用 read_draft 取得內容 → 按 4 面向審查 → 輸出結構化報告
審查面向
1. Structure：必要章節齊全、Abstract 結構化、邏輯順序（Major if 缺少/亂序）
2. Methods：
| 項目 | Major/Minor |
|---|---|
| Ethics/IRB statement | Major |
| Statistical methods | Major |
| Sample size justification | Minor |
| Primary outcome definition | Minor |
| Inclusion/exclusion criteria | Minor |
3. Statistics：
| 項目 | Major/Minor |
|---|---|
| Confidence intervals | Major |
| P-value reporting (精確值) | Minor |
| Multiple comparisons correction | Minor |
| Effect size | Minor |
4. Writing：
| 項目 | Major/Minor |
|---|---|
| Limitations section | Major |
| Causal language in observational | Major |
| Word count / Reference density | Minor |
判定
Major ≥3 → 🔴 Major revision required | 1-2 → 🟡 Major revision | 0 → ✅ Minor/Accept
Framework 2: Reporting Guideline Compliance
read_draft → 自動偵測/指定 guideline → 逐項檢查 → 合規報告
Auto-Detection
| 草稿特徵 | 指引 |
|---|---|
| randomized, RCT, blinded, placebo | CONSORT |
| systematic review, meta-analysis | PRISMA |
| case report, case presentation | CARE |
| observational, cohort, case-control | STROBE |
CONSORT (RCT, 21 items)
Title(1), Abstract(2), Intro(3-4: background, objectives), Methods(5-13: design, eligibility, interventions, outcomes, sample size, randomization, allocation concealment, blinding, statistics), Results(14-19: flow, recruitment, baseline, N analysed, outcomes+CI, harms), Discussion(20: interpretation+limitations), Other(21: registration)
STROBE (Observational, 16 items)
Title(1), Intro(2-3), Methods(4-11: design, setting, participants, variables, measurement, bias, sample size, statistics), Results(12-14: participants, descriptive, outcome), Discussion(15: interpretation), Other(16: funding)
PRISMA (SR, 17 items)
Title(1), Abstract(2), Intro(3), Methods(4-11: protocol, eligibility, sources, strategy, selection, extraction, RoB, synthesis), Results(12-15: flow, characteristics, results, heterogeneity), Discussion(16), Other(17: registration)
CARE (Case Report, 13 items)
Title(1), Abstract(2), Intro(3), Patient(4-7: demographics, complaints, history, exam), Clinical(8-10: diagnosis, intervention, follow-up), Discussion(11-12: strengths/limits, rationale), Patient(13: consent)
判定
≥80% → 🟢 Good | 60-80% → 🟡 Needs attention | <60% → 🔴 Significant gaps
使用原則
- 語意理解，非關鍵字匹配（「倫理委員會核准」= ethics approval）
- Major = 可能被 reject，Minor = 改善建議
- 給具體修復建議
- 多檔案支持（Introduction、Methods 可能在不同檔案）