convez.ai
Skill
Viprasol-Tech

onchain-forensics

Viprasol-Tech/onchain-forensics

Ver no GitHub
InicialPluginScore · de 0 a 100

AI agent skill for on-chain fund tracing & address-clustering investigations — MIH, change-address, FIFO mixer & bridge heuristics, OFAC checks, Daubert-grade evidence. Works with Claude Code, Codex, Cursor. By Viprasol Tech.

Instalação

$npx convez-ai Viprasol-Tech/onchain-forensics

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

On-Chain Forensics — Agent Skill

Trace fund flows across chains and cluster addresses with the standard 6-stage forensic workflow — MIH, change-address, FIFO mixer heuristics, bridge/DEX hops, OFAC screens, and a reproducible case report.

GitHub stars License: MIT Agent Skills Workflow Heuristics Version

Works with: Claude Code · OpenAI Codex · Cursor · GitHub Copilot · Gemini CLI · 26+ tools that support portable agent skills.


Quick Install

Clone the skill into your agent's skills directory:

# Claude Code
git clone https://github.com/Viprasol-Tech/onchain-forensics \
  ~/.claude/skills/onchain-forensics

# OpenAI Codex
git clone https://github.com/Viprasol-Tech/onchain-forensics \
  ~/.codex/skills/onchain-forensics

# Cursor / Copilot / Gemini CLI / other tools — clone anywhere your tool
# loads skills from, e.g.
git clone https://github.com/Viprasol-Tech/onchain-forensics \
  ~/.config/skills/onchain-forensics

The skill is a single self-contained skill.md (plus examples) — no dependencies, no install step, nothing to build.

Try It

Paste a few tx hashes, an address list, or a flow export and ask:

Trace these funds and tell me where they went — cluster the addresses and
flag any sanctioned-cluster contact.

The skill will scope the case, pick the right heuristics for the chain (UTXO vs account), walk the hops, call out mixer / bridge / DEX legs, score attribution confidence, and produce a reproducible investigator's case report.


Why This Exists

Modern crypto laundering chains 3–5 obfuscation techniques in sequence — peel chain → bridge → DEX → mixer → bridge — and a single missed clustering heuristic loses the trail. Professional investigators run the same 6-stage workflow every time: collect, cluster, attribute, graph, cross-chain hop, document. They label every claim with the heuristic used and its known limits so the report holds up under Daubert review. This skill encodes that workflow so your agent runs it the same way each time, picks the right heuristics for UTXO vs account-based chains, names confidence on every attribution, and never silently invents on-chain state.

What It Does

  1. Intake & scope — case type, chains, seed addresses/tx hashes, value at risk, timeframe, and the data the user can actually provide.
  2. Chooses heuristics by chain model — Multi-Input-Ownership and change-address for UTXO chains; deposit-address clustering, common gas funder, ENS reuse, and bridge-receipt matching for EVM/Solana.
  3. Walks the hops — direct transfer, DEX swap, bridge, mixer, peel chain, CEX deposit. Normalizes amounts in native units and USD where priced.
  4. Handles obfuscation chains — Tornado Cash with its OFAC sanction context and the published FIFO-temporal-matching defeat (5.1–12.6% baseline + 15–22pp from FIFO, per arXiv 2510.09433); Railgun and eXch flagged; bridges correlated by lock-mint event pairs.
  5. Attributes carefully — OSINT, public labels (Arkham, Nansen, Etherscan, Lookonchain), OFAC SDN, with Low / Med / High confidence and cited evidence on every row.
  6. Severity-tags red flags — 🔴 sanctioned contact / 🟠 mixer-adjacent / 🟡 multi-hop proximity / 🟢 / ℹ️.
  7. Outputs an investigator's case report — TL;DR + verdict, cluster table, hop-by-hop flow, mixer/bridge/DEX call-outs, attribution table, IOCs, reproducibility log (every step labeled with heuristic + source), and a Daubert-grounded disclaimer.

Features

  • 6-stage workflow — collect, cluster, attribute, graph, cross-chain, document. The Chainalysis-grade investigator's loop.
  • UTXO vs account-based — never confuses the two; MIH only fires on UTXO chains, deposit-address clustering only on account-based.
  • Mixer handling — Tornado Cash sanction context, founder-conviction note, and the published FIFO/address-reuse/gas-funder defeats with honest candidate-set caveats. Railgun and eXch flagged.
  • Bridge correlation — canonical lock-mint event pairing, AMM-bridge event-log scan, intent-solver caveat.
  • Severity ladder — 🔴 sanctioned / 🟠 mixer-adjacent or peel-chain / 🟡 multi-hop indirect / 🟢 / ℹ️.
  • Reproducibility log — every step in the report is labeled with the heuristic name + the source/link a human can re-walk to verify.
  • Daubert-aware disclaimer — testable / known error rates / peer-reviewed / generally accepted; the skill cites the standard but doesn't pretend to satisfy it alone.

Heuristic-confidence quick reference

HeuristicChain modelStrengthDefeat
Multi-Input Ownership (MIH)UTXOStrong (gold standard)CoinJoin / PayJoin / multi-sig
Change-address detectionUTXOMediumRandomized modern wallets
Deposit-address clusteringAccountStrong with hot-wallet labelsSelf-custody / DEX-only flows
Common gas funderAccountMediumExchange withdrawals fund many users
FIFO temporal matchingMixer+15–22pp over baseline (arXiv 2510.09433)Disciplined timing offsets
Bridge-receipt matchingCross-chainStrong on canonical bridgesIntent-based solvers, hop-aggregation

See full worked reports in examples/:

  • Stolen ETH trace — phishing victim, drainer EOA, peel chain into Tornado Cash, FIFO candidate set, CEX deposit attribution, 🟠 / 🔴 verdict.
  • BTC rug-pull cluster — MIH on the deposit side to cluster founder wallets, change-address chain into a peel, CEX deposit recognized via hot-wallet pattern, 🔴 verdict.

Not Legal Advice

This skill provides probabilistic investigative leads, not court-ready evidence. Clustering and mixer-deanonymization heuristics have known false-positive and false-negative rates; the skill cannot fetch on-chain state and reasons only over what you provide. For law-enforcement, civil litigation, or regulatory action, pair this with professional tooling (Chainalysis Reactor, TRM Labs, Elliptic) and subpoena-backed exchange KYC data, and have it reviewed by a qualified investigator and counsel. Not affiliated with or endorsed by Anthropic or any forensics vendor.


Contact — Viprasol Tech Private Limited

Viprasol Tech — fintech software, AI agents, algorithmic trading systems, and B2B SaaS.

License

MIT © 2025 Viprasol Tech Private Limited

Ver o original no GitHub

Bom para

Relacionados