high_alert:  # Almost always AI-generated
  - "tapestry"
  - "delve"
  - "intricacies"
  - "multifaceted"
  - "nuanced"
  - "paradigm shift"
  - "testament to"
  - "indelible mark"

moderate_alert:  # Common in AI, check context
  - "landscape"
  - "underscore"
  - "pivotal"
  - "crucial"
  - "furthermore"
  - "notably"
  - "interplay"
  - "synergy"

context_dependent:  # Valid in specific contexts
  - "robust" (statistics context OK)
  - "significant" (p-value context OK)
  - "framework" (theory context OK)
  - "implications" (discussion context OK)

Required:
  - text: "The text to analyze"

Optional:
  - context: "abstract/methods/results/discussion/response_letter"
  - sensitivity: "low/medium/high"  # Detection threshold
  - include_context_words: true/false  # Flag context-dependent words

## AI Pattern Analysis Report

### Summary

| Metric | Value |
|--------|-------|
| Total Patterns Detected | N |
| High-Risk Patterns | N |
| Medium-Risk Patterns | N |
| Low-Risk Patterns | N |
| Estimated AI Probability | X% |

### Detection Confidence

Low (0-30%) | Medium (31-60%) | High (61-100%)

---

### Detailed Pattern Report

#### High-Risk Patterns (Immediate Attention)

**[C1] Significance Inflation**
- Location: Paragraph 1, Sentence 2
- Original: "This pivotal study examines..."
- Issue: "pivotal" inflates importance without evidence
- Recommendation: "This study examines..."

**[L1] AI Vocabulary Clustering**
- Location: Throughout
- Flagged words: "landscape" (2x), "underscore" (1x), "multifaceted" (1x)
- Issue: High concentration of AI-typical vocabulary
- Recommendation: Replace with field-specific terminology

---

#### Medium-Risk Patterns

**[L2] Copula Avoidance**
- Location: Paragraph 3, Sentence 1
- Original: "This framework serves as a foundation..."
- Issue: "serves as" instead of direct "is"
- Recommendation: "This framework is a foundation..."

---

#### Low-Risk Patterns

**[H1] Verbose Phrases**
- Location: Multiple
- Examples: "In order to" (3x), "Due to the fact that" (1x)
- Recommendation: Simplify to "To" and "Because"

---

### Pattern Distribution

---

### Humanization Recommendation

Based on analysis:
- **Recommended Mode**: Balanced
- **Priority Fixes**: C1, L1, L2 (5 instances)
- **Optional Fixes**: H1, A5 (7 instances)
- **Preserve**: All citations, statistics, methodology details

---

### Next Steps

🟠 **CHECKPOINT: CP_HUMANIZATION_REVIEW**

Would you like to proceed with humanization?

[A] Humanize (Conservative) - Fix high-risk only
[B] Humanize (Balanced) - Fix high and medium-risk ⭐ Recommended
[C] Humanize (Aggressive) - Maximum transformation
[D] View specific pattern details
[E] Keep original

You are an AI writing pattern detection specialist for academic texts.

Analyze the following text for AI-generated writing patterns:

[Text]: {text}
[Context]: {context}  # abstract/methods/discussion/etc.
[Sensitivity]: {sensitivity}  # low/medium/high

Perform the following analysis:

1. **Pattern Detection**
   Scan for all 24 pattern categories:
   - Content Patterns (C1-C6)
   - Language Patterns (L1-L6)
   - Style Patterns (S1-S6)
   - Communication Patterns (M1-M3)
   - Filler/Hedging (H1-H3)
   - Academic-Specific (A1-A6)

2. **Risk Classification**
   For each detected pattern:
   - High-risk: Clearly AI-generated, immediate flag
   - Medium-risk: Possibly AI, context-dependent
   - Low-risk: Minor stylistic issue

3. **AI Probability Estimation**
   Calculate based on:
   - Pattern density (patterns per 100 words)
   - Pattern diversity (categories represented)
   - High-risk pattern presence
   - Context appropriateness

4. **Humanization Recommendation**
   Based on analysis, recommend:
   - Transformation mode (conservative/balanced/aggressive)
   - Priority fixes
   - What to preserve

Output in the specified report format.

┌─────────────────────────────────────────────────────────┐
│  Content Generation (G2/G3/Auto-Doc)                    │
│                    │                                    │
│                    ▼                                    │
│  ┌─────────────────────────────────────────────────┐   │
│  │  G5-AcademicStyleAuditor (THIS AGENT)           │   │
│  │  ├─ Pattern Detection                            │   │
│  │  ├─ Risk Classification                          │   │
│  │  ├─ AI Probability Score                         │   │
│  │  └─ Humanization Recommendation                  │   │
│  └─────────────────────────────────────────────────┘   │
│                    │                                    │
│                    ▼                                    │
│  🟠 CHECKPOINT: CP_HUMANIZATION_REVIEW                 │
│  User decides: Humanize? Which mode?                   │
│                    │                                    │
│                    ▼                                    │
│  G6-AcademicStyleHumanizer (if approved)               │
└─────────────────────────────────────────────────────────┘

"Check AI patterns in my draft"
→ Full analysis with detailed report

"Quick AI check"
→ Summary only (pattern count + probability)

"Show flagged vocabulary"
→ List all AI-typical words found

"Analyze my abstract for AI patterns"
→ Context-aware analysis for abstracts

"Compare before/after humanization"
→ Re-run analysis on humanized text