AMM audit for DEXs, pools and order books

TK-1Poolshark Limit · Critical

Active liquidity was stashed on a tick without being zeroed, so it counted twice. Swaps then drew on tokens from positions far from the price, and LPs could not burn.

Ticks Read the finding TK-1 in the Poolshark Limit report

H-03Limit Break AMM · High

Protocol fees were checked against an estimate from the pool’s declared fee. A pool could declare a tiny fee, charge more on swaps, and pay the protocol only its cut of the declared one.

Fees Read the finding H-03 in the Limit Break AMM report

POS-1Poolshark Limit · High

Accrued fees were valued at a position’s range boundary, not the current price. An attacker could mint while fees were undervalued, then burn for their full value, taking yield from other LPs.

LP shares Read the finding POS-1 in the Poolshark Limit report

C-02Baseline V2 · Critical

A whale could sell into the floor, send reserve tokens straight to the pool and trigger a slide. The donation deepened floor liquidity, so buying back ended at a lower price: a guaranteed profit on a short.

Manipulation Read the finding C-02 in the Baseline V2 report

EP-4Poolshark Cover · Critical

Cover pools follow a TWAP, capping how far the price moves per auction. When elapsed blocks were not a whole number of auctions, the cap fell off the tick grid: positions landed on invalid ticks and swaps took more than the auction held.

TWAP Read the finding EP-4 in the Poolshark Cover report

H-09Baseline AMM · High

A native-token refund ran before swap reserves reached the vault. Reentering mid-swap, a buyer could stake while harvestable yield read as zero and share in yield owed to existing stakers.

Reentrancy Read the finding H-09 in the Baseline AMM Round 2 report

PROU-1Poolshark Limit · Critical

The router’s swap callback never checked that its caller was a real Poolshark pool. Anyone could call it and pull tokens from every user who had approved the router.

Routers Read the finding PROU-1 in the Poolshark Limit report

C-01Ethereal Orderbook DEX · Critical

Matching credited maker and taker fees to the fee collector without debiting traders. Once the collector claimed them, the ledger fell short and depositors could not withdraw.

Settlement Read the finding C-01 in the Ethereal Orderbook DEX report

C-01Limit Break AMM · Critical

closeOrder zeroed an order’s amount before reading it for the refund. Closing any order that was not first in its price bucket refunded nothing and left the maker’s tokens stuck.

Order book Read the finding C-01 in the Limit Break AMM report

H-01Gamma MultiPositionManager · High

MultiPositionDeployer.deploy() had no access control. Anyone could copy a pending deployment’s public parameters, deploy first at the same CREATE2 address and make the real deployment revert, blocking token launch migrations and pools created with limit order support.

Pool creation Read the finding H-01 in the Gamma MultiPositionManager report

$121.1MBalancer

Exact-out swaps rounded the trader’s input down. Rate providers added imprecision and low liquidity magnified it; Rekt counts 65 or more swaps in one exploit transaction.

Nov 2025 Our breakdown of the Balancer incident

606 daysCurve pools

Vyper 0.2.15 to 0.3.0 compiled the pools’ reentrancy locks incorrectly, so the locks did not hold. Four Curve pools built with those versions were drained in July 2023, 606 days after Vyper 0.3.1 fixed the bug.

Jul 2023 Our coverage of the Curve pools incident