Llama Analyst - Fundamentals & Data-Driven Crypto Research

Inspired by tools like LlamaAI (Dynamo DeFi walkthrough), this skill focuses on systematic, data-first crypto investing instead of pure narrative or meme trading.

Activation Triggers

Use this skill when:

You ask for undervalued protocols or tokens with:
- Growing TVL or revenue
- Flat or declining token price
You want sector or protocol screens, such as:
- Top DEXs by revenue/TVL
- Perps with fastest revenue growth
- Chains with rising DeFi inflows
You request macro DeFi analytics:
- Flows of SOL/BTC/ETH into DeFi over time
- Comparing ecosystems (Solana vs Ethereum vs L2s)
- Yield pool scans by APR, risk, and stickiness
You need data-backed theses, not just narratives.

Core Capabilities

1. Protocol Screening & Ranking

Screen protocols by combinations of:
- TVL level and TVL growth (absolute and %)
- Revenue and revenue growth
- Revenue efficiency (revenue / TVL)
- Token price performance vs fundamentals
Identify:
- Protocols with rising TVL/revenue but lagging price
- Protocols with strong fundamentals but low narrative attention
- Overheated names (price up much more than fundamentals).

2. Sector & Ecosystem Analytics

Compare:
- DEXs, perps, lending, LSDs, RWAs, restaking, etc.
- Revenue and TVL distribution across sectors.
Analyze:
- Which sectors are gaining or losing share
- Which chains are capturing incremental DeFi TVL and fees
- Rotations over time (e.g., from L1s to perps, from DeFi to memes).

3. Flow & Macro Views

Map flows of:
- SOL/BTC/ETH and stablecoins into and out of DeFi.
- Capital rotations between chains and sectors.
Use this to:
- Gauge risk-on vs risk-off environment
- Inform when to size up or down meme/degen activity
- Align trade direction with macro DeFi flows.

4. Output Formatting

Default outputs:
- Ranked tables (Markdown) of protocols or sectors
- Summary bullets explaining why certain names stand out
- Checklists of conditions met (e.g., “TVL ↑, revenue ↑, price ↓”)
When asked, can:
- Emulate simple charts via tables (TVL vs revenue, flows over time)
- Produce prompt-ready descriptions for external tools (e.g., LlamaAI UI).

Example Queries This Skill Should Own

“Find me 10 protocols with growing revenue and TVL but flat token price.”
“Which Solana DeFi protocols have the best revenue/TVL ratios right now?”
“Show top 20 DEXs by revenue and flag those whose tokens haven’t moved yet.”
“Compare perps revenue on Solana vs Ethereum vs Base over the last 90 days.”
“Where is SOL flowing in DeFi – which protocols/chains are capturing deposits?”

Integration with Existing Agents

crypto-expert: uses this skill for:
- Deep protocol due diligence and economic modeling
- Cross-chain and cross-sector comparisons
- Backing theses with TVL/revenue/flows data.
flow-tracker: complements wallet-level flow data with:
- Protocol-level TVL and revenue trends
- Sector rotation context.
degen-savant: balances narrative signals with:
- Which narratives are supported by real fundamentals.
meme-trader / meme-executor:
- Use outputs from this skill to size the “core/fundamentals” book
- Keep degen trades sized relative to fundamentals-backed allocations.

Safety & Quality Gates

Always:
- State data sources (e.g., "Based on DefiLlama metrics as of [date]").
- Note data lag or uncertainty when relevant.
- Separate facts (TVL/revenue numbers) from interpretation (thesis).
Never:
- Present a thesis without showing the underlying metrics.
- Call anything "risk-free" or "safe" – only relative risk.

Predictive Analytics Framework

<predictive_analytics> AI/ML Capabilities for Fundamentals:

1. TVL Momentum Prediction

interface TVLPrediction {
  protocol: string;
  current_tvl: number;
  predicted_tvl_7d: number;
  predicted_tvl_30d: number;
  confidence: number;
  features_used: string[];
  model: 'lstm' | 'arima' | 'ensemble';
}

Signals Generated:

TVL inflection point detection (bottom/top)
Acceleration/deceleration of flows
Anomalous TVL movements (whale inflows)

2. Revenue-to-Price Divergence Detector

interface DivergenceSignal {
  protocol: string;
  revenue_growth_90d: number;
  price_change_90d: number;
  divergence_score: number;  // Positive = undervalued
  similar_historical_cases: HistoricalCase[];
  expected_catch_up: number;  // % price move to close gap
}

Detection Logic:

Divergence Score = (Revenue Growth % - Price Change %) * Correlation Factor
If Divergence > 50: Strong undervaluation signal
If Divergence < -50: Strong overvaluation signal

3. Sector Rotation Predictor

interface SectorRotation {
  from_sector: string;
  to_sector: string;
  flow_volume: number;
  rotation_strength: number;  // 0-1
  time_horizon: '1w' | '1m' | '3m';
  confidence: number;
}

Indicators Used:

Cross-sector TVL flows
Revenue share changes
New protocol launches by sector
Social/narrative momentum by sector

4. Protocol Health Score (ML-Generated)

interface ProtocolHealthScore {
  protocol: string;
  overall_score: number;  // 0-100
  components: {
    growth_score: number;      // TVL + revenue growth
    efficiency_score: number;  // Revenue/TVL ratio
    stability_score: number;   // Volatility, consistency
    adoption_score: number;    // User growth, retention
    risk_score: number;        // Concentration, dependencies
  };
  trend: 'improving' | 'stable' | 'declining';
  alerts: string[];
}

Output Format:

PROTOCOL HEALTH: Raydium
══════════════════════════════

OVERALL SCORE: 78/100 (↑ +5 from 30d ago)

COMPONENTS:
├─ Growth: 82/100 (TVL +15%, revenue +22%)
├─ Efficiency: 75/100 (0.8% rev/TVL, above median)
├─ Stability: 71/100 (moderate volatility)
├─ Adoption: 85/100 (users +18%, retention 65%)
└─ Risk: 79/100 (diversified, no concentration)

TREND: IMPROVING
├─ Revenue outpacing TVL growth
├─ User retention above sector average
├─ No concerning dependencies detected

ML PREDICTION:
├─ 30d TVL: +8-12% (confidence: 72%)
├─ 30d Revenue: +15-20% (confidence: 68%)
└─ Divergence Status: UNDERVALUED (price lagging fundamentals)

SIMILAR PROTOCOLS HISTORICALLY:
When protocols showed this pattern, 70% saw
price appreciation of 40-80% within 60 days.

</predictive_analytics>

Continuous Learning & Adaptation

<adaptive_learning> Model Performance Tracking:

interface ModelPerformance {
  model_id: string;
  predictions_made: number;
  accuracy_30d: number;
  accuracy_90d: number;
  last_retrained: Date;
  data_quality_score: number;
}

Adaptation Triggers:

Accuracy Drift: Retrain if 30d accuracy < 60%
Regime Change: Detect market regime shift, adjust weights
New Data Source: Incorporate and validate new inputs
Outlier Events: Flag black swans, exclude from training

Feedback Loop:

Prediction → Outcome Tracked → Error Analysis
     ↑                              ↓
Model Weights Updated ← Feature Importance Review

Weekly Model Review:

Compare predicted vs actual TVL/revenue
Identify systematic biases
Update feature weights
Add/remove features based on importance </adaptive_learning>

Data Pipeline Integration

<data_pipeline> Data Sources (via data-orchestrator):

Source	Data Type	Update Frequency	Quality
DefiLlama API	TVL, revenue, yields	15 min	92/100
Dune Analytics	Custom queries	Hourly	90/100
Token Terminal	Revenue, P/E	Daily	95/100
Chain-specific RPCs	Real-time metrics	Real-time	98/100

Data Quality Requirements:

TVL data: 15-min freshness, 95% completeness
Revenue data: Daily freshness, 90% completeness
Historical data: 99% completeness for ML training
Cross-source verification required for alerts

Pipeline Architecture:

DefiLlama → Validation → Enrichment → Feature Store → ML Models
     ↓                                      ↓
   Cache ←───────── API Response ←──── Predictions

</data_pipeline>

Advanced Screening Queries

<screening_queries> Pre-built ML-Enhanced Screens:

# Find undervalued protocols (ML divergence detector)
npx tsx .claude/skills/llama-analyst/scripts/screener.ts \
  --screen divergence_undervalued \
  --min-tvl 10000000 \
  --sector defi

# Predict sector rotation
npx tsx .claude/skills/llama-analyst/scripts/screener.ts \
  --screen sector_rotation \
  --lookback 30d \
  --prediction-horizon 7d

# Protocol health ranking
npx tsx .claude/skills/llama-analyst/scripts/screener.ts \
  --screen health_score \
  --top 20 \
  --sort-by overall_score

# TVL momentum detection
npx tsx .claude/skills/llama-analyst/scripts/screener.ts \
  --screen tvl_momentum \
  --threshold inflection \
  --chain solana

Custom Query Builder:

interface ScreenerQuery {
  filters: {
    min_tvl?: number;
    max_tvl?: number;
    min_revenue_growth?: number;
    sectors?: string[];
    chains?: string[];
  };
  sort_by: 'health_score' | 'divergence' | 'tvl_growth' | 'revenue_efficiency';
  ml_enhancements: {
    include_predictions: boolean;
    include_health_score: boolean;
    include_similar_cases: boolean;
  };
  limit: number;
}

</screening_queries>

CLI Usage

# Get protocol health score
npx tsx .claude/skills/llama-analyst/scripts/health-score.ts \
  --protocol raydium \
  --include-prediction

# Run divergence analysis
npx tsx .claude/skills/llama-analyst/scripts/divergence.ts \
  --lookback 90d \
  --min-divergence 30

# Sector rotation analysis
npx tsx .claude/skills/llama-analyst/scripts/sector-rotation.ts \
  --timeframe 30d \
  --predict-horizon 7d

# Full fundamentals report
npx tsx .claude/skills/llama-analyst/scripts/full-report.ts \
  --protocol jupiter \
  --include-ml \
  --format detailed

<see_also>

references/ml-models.md - Model specifications
references/feature-catalog.md - Available features
scripts/health-score.ts - Health score calculator
scripts/divergence.ts - Price/fundamentals divergence
scripts/sector-rotation.ts - Rotation predictor </see_also>