Adversarial academic text review with dual-agent critique loop. Triggers when user requests "review", "审查", "polish", or uses /review command. Uses Critic Agent to identify weaknesses, then Improver Agent to refine text. Iterates 2-3 rounds until convergence.
Improve academic writing through systematic adversarial critique. A Critic Agent identifies weaknesses without providing solutions; an Improver Agent refines the text based on critique. This loop repeats until stable.
User Text Input
│
▼
┌─────────────────┐
│ Critic Agent │ ← Identifies issues (no solutions)
│ (质疑者) │
└────────┬────────┘
│ Critique Report
▼
┌─────────────────┐
│ Improver Agent │ ← Refines text to fix issues
│ (改进者) │
└────────┬────────┘
│ Improved Text + Report
▼
┌─────────────┐
│ USER DECIDES │ ← Human approves or requests more rounds
└─────────────┘
│
┌─────┴─────┐
│Continue? │
└───────────┘
│Yes │No (stop)
▼ ▼
Next Round Final Text
(max 3)
Human Decision Points:
/review/planner or paper-outline-generator instead/planner firstBefore running the review loop:
The Critic Agent evaluates text and produces a structured critique:
Precision - Are claims stated precisely?
Support - Are claims backed by evidence?
Logic Flow - Do ideas connect logically?
Completeness - Is anything critical missing?
Academic Register - Is tone consistent and appropriate?
{
"critique": [
{
"issue": "Vague quantification",
"location": "Paragraph 2, sentence 3",
"text": "\"showed significant improvement\"",
"problem": "What makes it significant? Undefined without context or metrics."
}
],
"summary": "Main weaknesses identified: (1) vague quantification, (2) missing citations..."
}
The Improver Agent refines text based on critique:
{
"improvements": [
{
"original": "showed significant improvement",
"revised": "improved by 23.5% (p<0.01, n=100)",
"reason": "Quantified improvement and added statistical evidence"
}
],
"revised_text": "Full revised text here..."
}
Stopping Conditions (measurable):
Escalation Path: If Critic identifies structural/methodological issues beyond surface text polish (e.g., logical flaws, missing sections, unsupported claims that need new research), flag these as "ESCALATION" and recommend:
/planner to redesign the section structure## Academic Review Complete
**Rounds**: 2/3
**Issues Found**: 5
**Issues Resolved**: 5
### Key Changes
1. Added quantitative metrics to vague claims
2. Added 3 missing citations
3. Strengthened transition between paragraphs 3-4
4. Replaced 2 colloquial phrases with academic alternatives
### Escalated Issues (requires author attention)
- [List structural/methodological issues that need more than polish]
### Revised Text
[Full revised text with changes highlighted]
---
## Original vs Final (summary)
[Side-by-side comparison for major changes]
/review my introduction draft/review 帮我审查一下methodology部分/review and polish the abstract