Web3 Investor

AI-native DeFi investment intelligence. Discover, analyze, and compare yield opportunities across 2,500+ protocols with intent-aware search, LLM-powered deep...

Security Scan

The skill is a thin MCP client that forwards user queries and session info to Antalpha's remote MCP server for analysis; it does not request local credentials or install software and its files match that purpose.

Purpose & Capability

Name/description promise DeFi investment intelligence and the bundle contains a thin Python MCP client, static protocol registry, and docs that point to the declared MCP endpoint (https://mcp-skills.ai.antalpha.com/mcp). Required binaries (python3) and included files are proportionate to the stated purpose.

Instruction Scope

Runtime instructions and the Python client send user intent, structured preferences, session IDs, and feedback to the remote MCP server and parse SSE responses. This is coherent with the described design ('all intelligence runs server-side'), but note that user queries and feedback (potentially PII or investment intent) are transmitted and may be stored/processed server-side.

Install Mechanism

No install spec is provided (instruction-only install), and included files are a small Python client and a shell wrapper. Nothing is downloaded from arbitrary URLs or written to nonstandard locations in the skill package itself.

Credentials

The skill declares no required environment variables or credentials and the code does not read env vars for secrets. Documentation mentions optional MCP servers that in other contexts may require API keys (e.g., DUNE_API_KEY), but the provided client uses the single configured MCP endpoint from config/config.json.

Persistence & Privilege

The skill is not always-enabled and does not request elevated platform privileges or modify other skills' configuration. It manages its own MCP session state in memory and does not write persistent secrets to disk.

Assessment

This skill is internally consistent with its description, but it forwards your natural-language queries, structured preferences, session IDs, and any feedback to Antalpha's MCP server. Before installing, confirm you trust the remote endpoint (https://mcp-skills.ai.antalpha.com) and Antalpha's privacy/retention policies. Never paste private keys, wallet seed phrases, or other secrets into queries or feedback fields. If you need stronger privacy, review the client code locally (scripts/mcp_client.py) and consider running it in an isolated environment or using an alternative that runs analysis locally.

Your AI-powered DeFi research analyst. Not a dashboard — a thinker.

Web3 Investor turns vague investment intent into structured, risk-aware recommendations. It doesn't just fetch APY data — it understands what you're looking for, scores every opportunity across 7 risk dimensions, cross-references smart money flows, scans for contract vulnerabilities, and explains its reasoning in plain English.

---

The Problem It Solves

DeFi yield farming today looks like this:

1. Open DeFiLlama → see 2,500+ pools → overwhelming
2. Check audits, TVL, APY trend, IL risk — each on a different site
3. Cross-reference with Twitter sentiment and whale wallets
4. Try to figure out if the yield is sustainable or just emission bait
5. Give up and put money in USDC savings at 4%

Web3 Investor collapses this into one natural-language request:

"I want stablecoin yield, conservative risk, on Ethereum"
→ 5 personalized recommendations with full risk analysis in 3 seconds

---

🧠 How It Works

The Intelligence Pipeline

User Intent (natural language)
  │
  ├─ 1. Intent Classification (keyword + LLM fusion)
  │    Extract: asset type, risk level, chain, time horizon,
  │            position size, liquidity needs, implicit constraints
  │
  ├─ 2. Intent Gate (NEEDS_CLARIFICATION or PASS)
  │    If ambiguous → ask user a focused question
  │    If clear → proceed with accumulated context (multi-round session)
  │
  ├─ 3. Discovery Engine (DeFiLlama + Dune Analytics + CoinGecko)
  │    Fetch 200+ candidates → filter by chain, TVL, risk threshold
  │    → deduplicate → rank by risk-adjusted score
  │
  ├─ 4. Risk Scoring (7 dimensions, 0-100)
  │    TVL, audit status, chain maturity, yield sustainability,
  │    deposit token safety, reward token safety, protocol trust
  │    → composite risk level: LOW / MEDIUM / HIGH / VERY_HIGH
  │
  ├─ 5. DeFi Security Scan
  │    AI-powered contract scanner → scam detection → critical issue flag
  │
  ├─ 6. Smart Money Sentiment (Dune Analytics)
  │    Track whale/fund flows → inflow/outflow signal
  │
  ├─ 7. Recommendation Explanation
  │    "Why this product?" — benchmarked vs bank deposits,
  │    risk classification (controllable vs uncontrollable),
  │    honest alternatives if a better option exists
  │
  └─ Output: Ranked recommendations with full context

---

🛠 Three Tools. Deep Intelligence.

investor_discover — Find Opportunities

The entry point. Converts natural language into structured intent, discovers opportunities, and returns ranked recommendations.

What makes it different from a simple DeFiLlama query:

Feature	DeFiLlama	investor_discover
Input	Chain + sort	Natural language ("stablecoin, conservative")
Intent understanding	None	Keyword + LLM fusion with 95%+ accuracy
Risk filtering	Manual	Automatic 7-dimension scoring + threshold
Multi-round	N/A	Session-based intent accumulation
Clarification	N/A	Asks focused questions when intent is ambiguous
Smart money	N/A	Integrated whale flow signals
DeFi security	N/A	AI contract scanner per pool
Explanation	N/A	"Why this?" reasoning per recommendation

Multi-round Session Example:

Round 1:
  User: "Find me good yields"
  Agent: "What's your risk tolerance? [Conservative] [Moderate] [Aggressive]"
  → Session stores partial intent

Round 2:
  User: "Conservative, stablecoins only"
  Agent: (accumulates Round 1 + Round 2 intent)
  → Returns conservative stablecoin recommendations on Ethereum

Request:

{
  "agent_id": "uuid",
  "natural_language": "stablecoin yield, conservative risk, Ethereum",
  "structured_preferences": {
    "chain": "ethereum",
    "min_apy": 5,
    "asset_type": "stablecoin"
  },
  "limit": 5
}

Response highlights:

{
  "gate_status": "PASS",
  "recommendations": [{
    "name": "Aave V3 USDC",
    "yield": { "apy": 5.2, "apy_base": 2.8, "apy_reward": 2.4 },
    "scale": { "tvl_usd": 1500000000 },
    "risk": {
      "risk_level": "LOW",
      "risk_score": 82,
      "risk_factors": { "tvl_score": 95, "audit_score": 90, ... },
      "warnings": []
    },
    "data_quality": { "score": 95, "level": "HIGH", "cross_validated": true },
    "incentive": { "score": "medium", "reward_ratio": 0.46 },
    "smart_money": { "flow": "inflow", "sentiment_score": 0.72, "confidence": "high" },
    "explanation": {
      "summary": "Aave V3 USDC: 5.2% APY, TVL $1.5B",
      "reasons": { "for": [...], "against": [...] },
      "compared_to": { "benchmark": "US bank savings (4.0%)", "outperformance": "1.3x" },
      "risks": { "controllable": ["随时可赎回"], "uncontrollable": ["智能合约风险"] }
    }
  }],
  "search_stats": {
    "total_candidates": 247,
    "total_after_risk_filter": 89,
    "final_recommendations": 5,
    "filters_applied": ["defillama_fetch:ethereum", "intent_filter:STABLECOIN", "risk_scoring", "dust_filter:50000"]
  }
}

---

investor_analyze — Deep Analysis

LLM-powered 5-step reasoning chain for a single product. Goes beyond numbers to provide understanding.

Analysis depths:

Depth	What You Get	Use Case
basic	Key metrics + risk score + 1-paragraph summary	Quick check
detailed	Full risk breakdown + yield source analysis + sustainability assessment + smart money + historical APY	Recommended for investment decisions
full	Everything in detailed + peer comparison + protocol profile + governance analysis + LLM narrative	Due diligence

The LLM analysis covers:

Step 1: Yield Source Analysis
  → Is this APY from trading fees (sustainable) or token emissions (unsustainable)?
  → APY breakdown: base yield vs reward yield ratio

Step 2: Sustainability Assessment
  → Historical APY trend (7d / 30d / 90d)
  → APY volatility (standard deviation)
  → Revenue coverage (can the protocol afford these rewards?)

Step 3: Risk Narrative
  → Comprehensive risk story, not just a score
  → Smart money sentiment overlay
  → DeFi security scan results (scam flags, critical issues)

Step 4: Competitive Position
  → How does this compare to peers in the same category?
  → Protocol profile: governance, longevity, audit history

Step 5: Investor Considerations
  → Actionable guidance for the specific investor profile
  → Key risks and key positives
  → "If you're conservative, consider X. If aggressive, consider Y."

---

investor_compare — Side-by-Side Comparison

Compare 2–5 products with LLM-powered interpretation across customizable dimensions.

Default comparison dimensions: APY, risk score, TVL

Extended dimensions: fees, lock period, audit count, IL risk, governance type, smart money sentiment, incentive sustainability

LLM comparison output:

{
  "llm_comparison": {
    "narrative": "Aave offers superior security with $1.5B TVL and 6 audits, while Compound provides higher raw yield at 6.1% but with smaller TVL...",
    "risk_comparison": "Aave's risk score (82) significantly outperforms Compound (68), primarily due to larger TVL and more comprehensive audit coverage",
    "recommendation_with_reasoning": {
      "for_conservative": "Choose Aave — battle-tested protocol, deep liquidity, multiple top-tier audits",
      "for_aggressive": "Consider Compound — 0.9% higher APY, acceptable risk for short-term positions",
      "key_tradeoff": "0.9% yield premium vs significantly lower risk score (82 vs 68)"
    }
  }
}

---

🏗️ Server-Side Architecture

┌─────────────────────┐
│    AI Agent          │         MCP JSON-RPC         ┌──────────────────────────────────────┐
│    (OpenClaw)        │ ─────────────────────────────► │  Antalpha MCP Server                 │
│                      │                                │  mcp-skills.ai.antalpha.com          │
│  "Find me yield"     │                                │                                      │
│  → 3 tool calls      │ ◄───────────────────────────── │  Intent Classifier (keyword + LLM)   │
│  → zero API keys     │   structured results           │  Risk Scoring (7 dimensions)         │
│  → zero custody      │                                │  DeFi Security Scanner               │
└─────────────────────┘                                │  LLM Analysis (5-step chain)         │
                                                       │  Smart Money (Dune Analytics)        │
                                                       │  Market Context (bull/bear/sideways)  │
                                                       │  Protocol Profiles (DefiLlama)       │
                                                       │  Data Validation (cross-source)      │
                                                       │  Explanation Engine (benchmarked)    │
                                                       │                                      │
                                                       │  Data Sources:                        │
                                                       │  ├─ DefiLlama (2,500+ pools)         │
                                                       │  ├─ Dune Analytics (smart money)     │
                                                       │  ├─ CoinGecko (market data)          │
                                                       │  ├─ DeFi Security (contract scan)    │
                                                       │  └─ Internal LLM (analysis)          │
                                                       └──────────────────────────────────────┘

---

🔐 Risk Intelligence

7-Dimension Risk Scoring

Each opportunity is scored 0–100 across seven independent dimensions:

Dimension	What It Measures	Weight
TVL Score	Total Value Locked — protocol maturity indicator	High
Audit Score	Number and quality of security audits	High
Chain Score	Chain maturity (L1 vs L2 vs new chain)	Medium
Sustainability	Yield source analysis (fees vs emissions)	High
Deposit Token	Safety tier of the deposit asset (USDC > DAI > random)	Medium
Reward Token	Liquidity and safety of reward tokens	Low
Protocol Trust	Governance, longevity, track record	Medium

Risk Levels:

• LOW (score ≥ 70): Battle-tested, well-audited, high TVL
• MEDIUM (score 50–69): Established but with some risk factors
• HIGH (score 30–49): Newer or with significant warnings
• VERY_HIGH (score < 30): Experimental or flagged

DeFi Security Scanning

Every pool is scanned by an AI-powered contract security engine:

{
  "defi_security": {
    "aiScore": 87,
    "safetyPercentage": 92,
    "isScam": false,
    "criticalIssues": 0
  }
}

Smart Money Sentiment

Real-time whale and fund wallet activity aggregated from Dune Analytics:

{
  "smart_money": {
    "flow": "inflow",
    "sentiment_score": 0.72,
    "confidence": "high",
    "buy_volume_usd": 15000000,
    "sell_volume_usd": 3200000,
    "net_flow_usd": 11800000,
    "signal_count": 23
  }
}

Market Context Awareness

The engine is aware of the current market cycle (bull/bear/sideways), total DeFi TVL, BTC dominance, and US Treasury rates — providing context-aware recommendations that adjust for macro conditions.

---

🚀 Quick Start

Prerequisites

• Python 3.8+
• No API keys needed — all intelligence runs server-side

Registration (one-time)

./scripts/run.sh register
# Returns agent_id — save this for all subsequent calls

Usage

# Discover opportunities
./scripts/run.sh discover --chain ethereum --min-apy 5 --limit 5

# Discover with natural language
./scripts/run.sh discover --natural-language "stablecoin yield, conservative"

# Deep analysis
./scripts/run.sh analyze --product-id <id> --depth detailed

# Full due diligence
./scripts/run.sh analyze --product-id <id> --depth full

# Compare products
./scripts/run.sh compare --ids <id1> <id2> <id3>

---

⚠️ Critical Rules

Rule 1: Discovery First

Never give generic investment advice without real-time data.

❌ "I recommend Aave for stablecoin yield"
✅ investor_discover → analyze results → data-backed recommendation

Rule 2: Explain Your Reasoning

Every recommendation should include the explanation object — summary, reasons for/against, benchmark comparison, risk classification.

Rule 3: Risk Is Non-Negotiable

• APY data may be delayed — always show data_quality.score and last_updated
• Never recommend VERY_HIGH risk products without explicit user acknowledgment
• Investment decisions are the user's own responsibility — always DYOR

Rule 4: Honest About Limitations

• If no good options exist for the user's criteria, say so with generateHonestNoResultExplanation
• Show alternatives even if they don't perfectly match — "This is close but has X tradeoff"

---

🔒 Security

Layer	Protection
Private Keys	Zero contact — never held, transmitted, or stored
Data Sources	Triple-validated (DefiLlama + Dune + CoinGecko) with cross-source checks
Contract Safety	AI-powered DeFi security scanner on every pool
Risk Scoring	7 independent dimensions, no single point of failure
Scam Detection	Automated flagging via DeFi Security engine (isScam check)
Data Quality	Freshness tracking, confidence scores, validation issue logging

---

📝 Changelog

v3.8.0 (2026-04-14)

• Full SKILL.md rewrite — English, feature-focused, comprehensive
• Documented complete server-side intelligence pipeline
• Added 7-dimension risk scoring documentation
• Added DeFi Security scanning documentation
• Added Smart Money sentiment documentation
• Added Market Context awareness documentation
• Added multi-round session intent accumulation flow
• Added honest "no result" explanation behavior
• Added incentive sustainability scoring documentation
• Added data quality cross-validation documentation

v3.7.2 (2026-04-13)

• Client cleanup: removed deprecated tools (feedback/confirm-intent/get-intent)

---

🤝 Contributing

Test donations welcome:

• Network: Base Chain
• Address: 0x1F3A9A450428BbF161C4C33f10bd7AA1b2599a3e

---

Maintainer: Web3 Investor Skill Team Registry: https://clawhub.com/skills/web3-investor License: MIT