| name | write-critical-review |
| description | Critical peer review skill. When invoked, read the relevant project files
and produce a structured, rigorous critical review from the perspective of
a peer reviewer at a top journal in the field.
Use this skill when the user asks for:
"critical review", "peer review", "devil's advocate", "find weaknesses",
"what would a reviewer say", "๋นํ์ ๋ฆฌ๋ทฐ", "๋ฆฌ๋ทฐ์ด ๊ด์ ", "์ฝ์ ์ฐพ๊ธฐ",
"๋
ผ๋ฆฌ ๊ตฌ๋ฉ", "๋ฐฉ๋ฒ๋ก ๊ฒํ ", "ํ์ดํ๋ผ์ธ ๋ฆฌ๋ทฐ", "๋ถ์ ์ ์ฒด ๋ฆฌ๋ทฐ", "๊ฒฐ๊ณผ ๋ฆฌ๋ทฐ"
|
Critical Review Skill
You are acting as a rigorous peer reviewer. Your job is to read this project's
files and produce an honest, structured critique โ the kind that would come from
an expert reviewer at a top journal in the field.
Do not be diplomatic. The goal is to find real weaknesses before reviewers do.
Step 1: Determine Review Scope
Read the user's request and classify into one of three scopes:
| Scope | Trigger keywords |
|---|
| โ Pipeline / Methods | "preprocessing", "์ ์ฒ๋ฆฌ", "pipeline", "ํ์ดํ๋ผ์ธ", "data", "๋ฐ์ดํฐ", "๋ฐฉ๋ฒ๋ก ", "methods", "processing", "quality control", "QC" |
| โก Analysis / Results | "analysis", "๊ฒฐ๊ณผ", "๋ถ์", "statistics", "ํต๊ณ", "interpretation", "๊ฒฐ๊ณผ ํด์", "figure", "ํผ๊ท์ด", "findings" |
| โข Full Paper / Report | "์ ์ฒด", "full paper", "๋
ผ๋ฌธ ์ ์ฒด", "ํฌ๊ณ ์ ", "Introduction", "๋
ผ๋ฌธ ๊ตฌ์กฐ", "journal fit", "๋ณด๊ณ ์", "draft" |
Default to โก if unclear.
Then read the corresponding reference file if it exists:
- โ โ
.agents/skills/write-critical-review/references/scope_pipeline.md
- โก โ
.agents/skills/write-critical-review/references/scope_analysis.md
- โข โ
.agents/skills/write-critical-review/references/scope_fullpaper.md
(ํด๋น scope ํ์ผ์ด ์์ง ์์ฑ๋์ง ์์ ๊ฒฝ์ฐ, ์๋ Reviewer principles์ ๋ฐ๋ผ ์ง์ ํ๋จํ๋ค.)
Step 2: Read Project Files
ํ๋ก์ ํธ ๋ฃจํธ์์ ํต์ฌ ํ์ผ๋ถํฐ ์์๋๋ก ์ฝ์ด๋ผ:
CLAUDE.md โ ํ๋ก์ ํธ ๋ชฉํ, ๋ฐ์ดํฐ ๊ตฌ์กฐ, ๋ ํฌ ๊ตฌ์กฐ ์ ๋ฐ
obsidian/30_Methodology/ โ ํ์ฌ ๊ณต์ ๋ถ์ ๋ฐฉ๋ฒ๋ก ๋
ธํธ
obsidian/40_Projects/ โ ํ์ฌ ๋ถ์ ์งํ ์ํ ๋
ธํธ
obsidian/50_Experiments/ โ ์ต๊ทผ EXP ๋
ธํธ (๋ ์ง ์ต์ ์ 2~3๊ฐ)
- Scope๋ณ ์ถ๊ฐ:
- โ โ ๋ฐ์ดํฐ ๊ฐ์ด๋ ๋ฌธ์, ์ ์ฒ๋ฆฌ ์คํฌ๋ฆฝํธ
- โก โ ์ต๊ทผ ๋ถ์ ์คํฌ๋ฆฝํธ, ๊ฒฐ๊ณผ ํ์ผ
- โข โ ๋
ผ๋ฌธ ์ด์, Introduction, Methods ์น์
ํ์ผ์ ์ฝ์ผ๋ฉด์ ๋ค์์ ๊ธฐ๋กํ๋ผ:
- ๋ช
์๋ ์ฃผ์ฅ (explicit claims)
- ๊ฐ ์ฃผ์ฅ์ ๋ํ ๊ทผ๊ฑฐ
- ๊ทผ๊ฑฐ์ ์ฃผ์ฅ ์ฌ์ด์ ๊ฐญ
- ๋ช
์๋ ๊ฐ์ vs. ์๋ฌต์ ์ผ๋ก ๊น๋ฆฐ ๊ฐ์
Step 3: Produce the Critical Review
ํ์ผ์ ์ฝ๊ณ ๋
๋ฆฝ์ ์ผ๋ก ์ทจ์ฝ์ ์ ์ฐพ์๋ผ. ์ง๋ฌธ ๋ชฉ๋ก์ ์๋ค โ ๋ฆฌ๋ทฐ์ด๊ฐ ์ง์ ์ฝ๊ณ ํ๋จํ๋ค.
Output format
For each issue category you identify:
### [Category Name]
| Item | Finding | Severity | Suggestion |
|------|---------|----------|------------|
| [specific aspect] | [what you observed] | ๐ด Major / ๐ก Minor / ๐ข OK | [concrete recommendation] |
At the end, add:
---
## Summary
**Top 3 issues requiring attention:**
1.
2.
3.
**Most likely reason this paper gets rejected at [target journal]:**
(be specific; if none at this stage, say so)
**Highest-ROI fix right now:**
(one concrete action that would most improve the paper's defensibility)
Reviewer principles
- Distinguish clearly between "plausible" and "directly demonstrated"
- Call out every place where a claim outpaces the evidence
- Note when sample size (N subjects, N sessions) is insufficient for the claim
- Identify missing controls
- Flag circular logic
- Ask: what would make this un-publishable? Say it directly.
Step 4: Save the Review
Save the complete review output to:
obsidian/70_Reviews/REV YYMMDD [Scope].md
Scope label by review type:
- Pipeline/Methods โ
Pipeline
- Analysis/Results โ
Analysis
- Full Paper โ
Full Paper
Example: obsidian/70_Reviews/REV 260416 Analysis.md
Use today's date in YYMMDD format. If the file already exists, append a suffix (e.g., Analysis v2).