Security audit summary

Full contract-level audit report: our internal security audit notes.

Contract-by-contract breakdown: see Contracts · Audit.

This page is the security-program landing — internal audit, external audit, bug bounty, responsible disclosure, and incident response.

Internal audit — 2026-05-18

• Auditor: Claude (Opus 4.7), internal pre-mainnet review
• Scope: 5 contracts + interfaces + MockClanker, ~600 LOC business logic
• Result: 0 critical, 3 high, 7 medium, 9 low/informational

The audit is read-only — no contract source was modified during the review. Findings are tracked in the roadmap’s Phase 5.2 remediation.

High-severity findings (block mainnet)

• H-01 Vote-state leak across rounds — voters from a disputed round are locked out of subsequent rounds.
• H-02 No bounty-token whitelist — fee-on-transfer / rebasing / blacklist tokens brick accounting.
• H-03 1-wei AGAINST short-circuit on finalize — a single bonder can decide every bounty pre-deadline.

External audit — pending

Engagement opens after the high-severity remediations land (Phase 5.2). Auditor candidates: see Roadmap.

Required scope:

1. Math precision proofs on BondingEscrow.claim across input grids.
2. Clanker v4 integration semantics on Base mainnet.
3. Front-running of finalize (MEV sandwich analysis).
4. Cross-bounty griefing.
5. Replay protection on dispute.
6. Storage packing of BountyState.

Bug bounty — pending

Live concurrently with the external audit. Scope and rewards: see Bug Bounty.

Responsible disclosure

Coordinated disclosure policy: see Responsible Disclosure.

For active exploits or imminent loss-of-funds situations, see Emergency Pause.

Security guarantees (current state)

What we currently guarantee:

• No custodial keys: the protocol never holds user EVM private keys. The relayer EOA only has permission to call commitChamber and deployIdea; it cannot move user funds.
• markGraduated authorization: callable only by the idea token itself or the factory, re-asserted on every PR touching ChamberRegistry.
• LP locks: every idea token’s LP is locked in ClankerLpLocker until year 2100. No rug vector at the LP level.
• Reentrancy: ReentrancyGuard on every state-mutating external entry point that moves tokens. SafeERC20 everywhere.
• No unchecked, no assembly: arithmetic safety governed by Math.mulDiv and Solidity 0.8 built-ins.

What we explicitly do not yet guarantee:

• Mainnet readiness: gated on Phase 5.2 + external audit. Sepolia only.
• DoS resistance on finalize: H-03 means review windows are short-circuit-able with 1 wei.
• Token-shape safety: H-02 means malicious ERC-20s can grief their own bounty.
• Bounded dust: M-01 means many-winner bounties leave unrecoverable residual.

What an attacker would target

Threat model summary:

The two material risks gating mainnet are H-02 and H-03. Both have clean fixes pending.

Threat modeling reference

• OWASP Top 10 for Smart Contracts (2023) 
• SWC Registry  — the canonical smart-contract weakness catalogue
• Trail of Bits’ SCSVS-checklist 

The audit explicitly cross-referenced every SWC entry. No matches outside the documented HIGH / MEDIUM findings.