Skip to content

tokamak-network/Argus

Repository files navigation

Argus

CI License: MIT OR Apache-2.0

The Hundred-Eyed Guardian for Ethereum

Real-time attack detection, post-hack forensics, and time-travel debugging for EVM transactions.

Existing security tools (Slither, Mythril, Echidna) analyze contracts before deployment. Argus protects after deployment — detecting attacks as they happen and analyzing them when they've already occurred.


What Argus Does

All features below have been implemented and tested in synthetic/demo environments. Production validation is in progress — see Roadmap.

Sentinel — Real-Time Attack Detection

A 2-stage detection pipeline integrated at the block processing level:

  • Pre-filter: Receipt-based heuristics (~10-50μs/tx) eliminate 99% of benign transactions
  • Deep analyzer: Full opcode-level replay on suspicious transactions only
  • Mempool monitoring: Detect attacks before they're included in a block
  • Auto-pause circuit breaker: Automatically halt processing on critical alerts
  • Multi-channel alerts: JSONL, Webhook (Slack/Discord/PagerDuty), WebSocket, Prometheus

Autopsy Lab — Post-Hack Forensics

Replay any mainnet transaction and generate a forensic report:

  • Attack pattern classification: Reentrancy, flash loans, price oracle manipulation, access control bypasses
  • Fund flow tracing: Track ETH and ERC-20 token movements through the attack
  • Markdown report generation: Structured timeline with attack phases, evidence, and suggested fixes

Time-Travel Debugger

GDB-style interactive replay at opcode granularity:

  • Forward/backward stepping through execution
  • Breakpoints on opcode, address, or storage slot
  • Full state inspection at any point in execution
  • Sub-50ms step navigation

Quick Start

Run the Sentinel demo

git clone https://github.com/tokamak-network/Argus.git
cd Argus
cargo run --example sentinel_realtime_demo

This simulates a block containing both benign and malicious transactions, showing Sentinel's detection pipeline in action:

Demo 1  Multi-TX Block Scanning
  TX 0: Simple ETH transfer (21k gas, success)
  TX 1: Flash loan via Aave (2.5M gas, 6 ERC-20 transfers)
  TX 2: 5 ETH transfer, reverted (950k gas)

  Alert #2:
    Priority: High
    Score:    0.75
    Summary:  Pre-filter alert: high-value-revert, unusual-gas, self-destruct
    Reason:   HighValueWithRevert { value_wei: 5000000000000000000 }

Demo 3  Mempool Pre-Filter (Pending TX Scanning)
  Mempool TXs scanned:  4
  Mempool TXs flagged:  3

Demo 4  Auto-Pause Circuit Breaker
  Critical alert → block processing HALTED

Run the Autopsy demo

cargo run --example reentrancy_demo

Deploys a vulnerable contract, executes a reentrancy attack, and generates a full forensic report:

Phase 1  Deploy & Execute
  Execution: SUCCESS (gas_used=82107)
  Opcode steps recorded: 80

Phase 2  Verify Attack
  Max call depth: 4  (need >= 3 for reentrancy)
  CALL opcodes:   4
  SSTORE opcodes: 4  (attacker counter writes)
  Result: CONFIRMED — reentrancy pattern detected

Phase 3  AttackClassifier
  Reentrancy (target=0x...0043)
    confidence: 90.0%
    evidence: Re-entrant call at step 47
    evidence: State modified at step 69
    evidence: Value transfer during re-entry

Phase 5  SentinelService Pipeline
  Alert Priority: Critical
  Score: 0.85
  Summary: Likely reentrancy attack (confidence: 90%)

Case Studies — Retroactive Analysis

Note: These are retroactive analyses of past exploits, not real-time detections. Argus has no production detection track record yet. We include these to demonstrate the detection logic, not to claim operational results.

Retroactive Analysis: $128M Balancer V2 Exploit

On November 3, 2025, an attacker exploited a rounding error in Balancer V2's batchSwap to drain $128M across 6 chains in under 30 minutes. We analyzed this transaction through Argus's detection pipeline:

  1. Pre-filter: Flags the transaction pattern (3M gas + 4.2KB calldata + Balancer Vault interaction)
  2. Deep Analyzer: Classifies as price manipulation (82% confidence) after opcode replay
  3. Autopsy: Generates a forensic report with fund flow tracing

Read the full analysis: docs/analysis-balancer-v2-exploit.md

Retroactive Analysis: $1.5B Bybit Exploit

On February 21, 2025, North Korea's Lazarus Group executed the largest crypto theft in history — $1.5B drained from Bybit's cold wallet via a supply chain attack on Safe{Wallet}'s front-end.

  1. Pre-filter: Flags the transaction for unusual DELEGATECALL to an unverified contract
  2. Deep Analyzer: Classifies as access control bypass (95% confidence) — proxy implementation overwritten
  3. Autopsy: Traces fund flow across 40+ intermediary wallets

Read the full analysis: docs/analysis-bybit-1.4b-exploit.md


How It Compares

Argus Forta Phalcon Tenderly Hexagate
Runtime detection Yes Yes (bot network) Yes Partial (alerts) Yes
Mempool pre-detection Yes No Yes Yes Yes
Post-hack forensics Yes No No Partial No
Open source Fully Partial No No No
Self-hosted Yes No (SaaS) No (SaaS) No (SaaS) No (SaaS)
Multi-chain No Yes (7+) Yes Yes (109) Yes
Production track record None yet 270M+ TX scanned 20+ hacks blocked 1.4M+ simulations Undisclosed

Argus is early-stage. Its primary differentiator today is being fully open-source and self-hostable. See competitive analysis for an honest, detailed comparison.


Architecture

                    ┌─────────────────────────────────────┐
                    │          Ethereum Network            │
                    └──────────────┬──────────────────────┘
                                   │
                    ┌──────────────▼──────────────────────┐
                    │           Mempool Monitor            │
                    │     (pre-execution calldata scan)    │
                    └──────────────┬──────────────────────┘
                                   │
              ┌────────────────────▼────────────────────────┐
              │              Sentinel Pipeline               │
              │                                              │
              │  ┌──────────┐    ┌────────────────────────┐ │
              │  │Pre-filter │───▶│    Deep Analyzer        │ │
              │  │ ~10-50μs  │    │ (opcode-level replay)  │ │
              │  │  per tx   │    │                        │ │
              │  └──────────┘    └───────────┬────────────┘ │
              └──────────────────────────────┼──────────────┘
                                             │
                    ┌────────────────────────▼─────────┐
                    │         Alert Dispatcher          │
                    │  JSONL / Webhook / WS / Prometheus │
                    └────────────────────────┬─────────┘
                                             │
                    ┌────────────────────────▼─────────┐
                    │      Auto-Pause Circuit Breaker   │
                    │   (halt on Critical severity)     │
                    └──────────────────────────────────┘

Features

Feature Description Feature Flag
Sentinel Real-time attack detection pipeline sentinel (default)
Autopsy Lab Post-hack forensic analysis autopsy (default)
Time-Travel Debugger Interactive opcode replay always included
CLI Interactive debugger shell cli
Dashboard Web UI for Sentinel metrics sentinel

Building

# Default (Sentinel + Autopsy)
cargo build

# With CLI debugger
cargo build --features cli

# All features
cargo build --all-features

Requirements: Rust 1.85+ (edition 2024)

Docker

Docker Hub image is not yet published. For now, build locally:

# Build locally
docker build -t argus-demo .

# Run the Sentinel demo
docker run argus-demo

# Run the Autopsy demo
docker run argus-demo reentrancy_demo

# Run the Dashboard demo
docker run argus-demo sentinel_dashboard_demo

Powered By

Argus uses ethrex LEVM as its EVM execution engine — a minimal, fast Ethereum Virtual Machine implementation in Rust. Argus depends on Tokamak Network's fork which includes the tokamak-debugger feature.


License

Dual-licensed under MIT or Apache 2.0.

Built by Tokamak Network.

About

Real-time Ethereum attack detection and post-hack forensics toolkit

Topics

Resources

License

Unknown, MIT licenses found

Licenses found

Unknown
LICENSE-APACHE
MIT
LICENSE-MIT

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages