Research · Web3 Security · High
Seed packs could bypass the VRF commitment check
A zero seed_slot skipped the commitment validation, allowing seed-pack purchase or opening to proceed before randomness had been committed.
8 November 2026
What went wrong
The seed-pack flow in seeds.rs::purchase_seed_pack did not reject a randomness account whose seed_slot was zero. That value indicated that the VRF commitment phase had not taken place.
The slot validation ran only when seed_slot was non-zero. As a result, an uncommitted account skipped the validation rather than causing the purchase or opening transaction to fail.
Why it mattered
The operation could proceed without the commitment expected by the commit-reveal flow. The finding describes the resulting risk as seed generation that was not backed by the intended randomness process, potentially allowing manipulation or predictability.
The missing condition was a prerequisite for advancing the operation. A transaction succeeding did not establish that randomness had been requested and committed as required.
The public report includes a purchase-without-commit test to illustrate the missing prerequisite. The relevant security property is that the operation must reject an account that has not entered the commitment phase.
The fix
The report recommends rejecting a zero seed_slot and enforcing the expected commit-slot timing before allowing the operation to proceed. Its proposed alternatives are to make the slot check unconditional or to add an explicit error branch for the uncommitted case.
This makes absence of a commitment a failing condition rather than a path around the timing check.
Takeaway
Review default and sentinel values as explicit states in a commit-reveal workflow. A check that validates only populated fields can let an uninitialized value bypass the prerequisite it was meant to enforce.
Shipping something similar? We review the code before it holds money.
Request an audit