DeFi On-Chain Analytics

An AI agent skill for conducting DeFi on-chain analysis using direct JSON-RPC calls. No APIs, no indexers — raw RPC to verifiable insights.

Install

npx skills add Omnis-Labs/defi-onchain-analytics

Or manually:

git clone https://github.com/Omnis-Labs/defi-onchain-analytics.git ~/.claude/skills/defi-onchain-analytics

What This Skill Does

Guides AI agents through a looped workflow for on-chain DeFi analysis — supporting scope amendments when discoveries change the question:

1. Scoping Gate — Define objectives, unit of analysis, anchor policy, decision dependencies
2. Discovery — Contract classification, proxy resolution, interface recovery (first-class step)
3. Data Collection — Batch reads, event log scanning, adaptive chunking, traces, bulk scaffolds
4. Interpretation — Classification-first analysis with multi-pass adversarial review
5. Rescope Decision — New question emerged? Lightweight re-entry to Phase 0. Otherwise proceed.
6. Sanity Check — Cross-validation, blind spot audit, gap logging
7. Synthesis — Mode-specific output profiles (forensic, due diligence, monitoring, etc.)

Capabilities

• Wallet profiling — Balance snapshots, transfer history, entity clustering, funding trace
• Protocol analysis — TVL decomposition, admin risk, oracle health, governance participation
• DEX analytics — LP position analysis, owner resolution, market structure, bot detection
• CLAMM vault analytics — Rebalance decomposition, fee separation, share-price time series, HODL benchmarks
• Token metrics — Supply audit, holder concentration, vesting schedules
• Contract inspection — Proxy detection, storage layout, event decoding, ABI resolution

Supported Chains

Chain	Chain ID	Public RPC Endpoints	Benchmark Date
Ethereum	1	24	2026-03-21
Arbitrum One	42161	14	2026-03-21
Base	8453	22	2026-03-21
BSC	56	26	2026-03-21
Polygon PoS	137	14	2026-03-21
Katana	747474	5	2026-03-21

108 verified endpoints with tier rankings, latency data, and getLogs support markers.

Skill Architecture

SKILL.md                    Main looped workflow (396 lines)
├── references/
│   ├── rpc-field-guide.md      RPC methods, chunking, L2 guide, explorer APIs
│   ├── rpc-endpoints.ts        108 verified endpoints across 6 chains
│   ├── common-abis.md          Objective router for ABI loading
│   ├── abis-core-tokens-vaults.md  ERC-20/721/1155/4626 signatures and selectors
│   ├── abis-dex-v3-v4-clamm.md     Uniswap V3/V4 and Algebra CLAMM state/events
│   ├── abis-proxy-and-multicall.md EIP-1967 proxy slots and Multicall3
│   ├── abi-fetching.md         Proxy detection and ABI resolution
│   ├── scoping-guide.md        Phase 0 detailed consultation guide
│   ├── investigation-discipline.md  7-layer anti-shortcutting defense
│   ├── data-collection-scaffold.ts  Bulk RPC with rate limiting, endpoint rotation, checkpoint/resume
│   └── proxy-resolver-scaffold.ts   Automated proxy chain resolution and selector extraction
└── patterns/
    ├── wallet-analytics.md     Address clustering, funding trace, Sybil detection
    ├── protocol-analytics.md   TVL, lending health, oracle monitoring
    ├── token-analytics.md      Supply audit, holder analysis
    ├── dex-analytics.md        V3 math, LP resolution, bot signals
    ├── clamm-vault-analytics.md  Rebalance decomposition, fee separation, share-price, LVR
    └── contract-inspection.md  Storage layout, proxy patterns, event decoding

Key Features

4-Tier Data Confidence System

Tier	Tag	Requires	Free Public RPC?
A	[CORE]	Standard JSON-RPC	Yes
B	[ARCHIVE]	Historical state >128 blocks	Rarely
C	[TRACE]	debug/trace namespace	No
D	[ENRICH]	External source (Etherscan, Sourcify)	Yes (not RPC)

Every finding is tagged with its data source tier. Unavailable tiers are disclosed, never silently skipped.

7-Layer Investigation Discipline

Prevents AI agents from taking analytical shortcuts:

1. Anti-Rationalization — Banned dismissal phrases that trigger deeper investigation
2. Iterative Depth — Multi-pass analysis with adversarial re-examination
3. Anti-Normalization — "Looks normal" is sophistication, not innocence
4. Blind Spot Audit — Mandatory disclosure of what was NOT investigated
5. Confidence Deepening — Low confidence on significant findings → must dig deeper
6. Adversarial Self-Review — Devil's Advocate questions per finding
7. Gap Logging — Every skipped method logged with impact

Adaptive RPC Chunking

Production-ready eth_getLogs chunking with:

• Per-provider block range limit table (Alchemy, Infura, dRPC, etc.)
• Error code recognition (-32005, -32602, -32614)
• TypeScript + viem code template with bisection
• Block explorer API fallback (Blockscout, Etherscan)

Protocol Coverage

• Uniswap V3/V4 — Full ABI reference, math formulas, position analytics
• Algebra CLAMM — Delta vs V3 (Camelot, QuickSwap, Katana DEX)
• ERC-4626 Vaults — Share/asset conversion, deposit/withdraw analysis
• L2 Chains — Block cadence, finality semantics, bridge tracing

Compatible Agents

Works with any agent that supports the skills format:

• Claude Code
• OpenCode
• Cursor
• Cline
• Windsurf
• And more on skills.sh

Contributing

Contributions welcome. The most impactful contributions are:

• New chain endpoints — Add verified RPC endpoints to references/rpc-endpoints.ts
• Pattern expansions — Extend analytical methods in patterns/ files
• ABI references — Add protocol ABIs to the appropriate split file in references/ (abis-core-tokens-vaults.md, abis-dex-v3-v4-clamm.md, or abis-proxy-and-multicall.md)
• Bug reports — Real-world cases where the skill gave incorrect guidance

License

MIT