Solidity Security Checklist: 75 Checks Before You Ship
Part of our guide to auditing a smart contract

Quick answer. No list covers every system, but this Solidity security checklist gives a 75-check pre-audit baseline across access control, external calls, math, integrations and upgrades. Each check names a finding category and the evidence to attach before reviewers start.
Video transcript
No checklist covers every system. This one gives you a seventy five check pre-audit baseline, and each check is tied to a finding category and the evidence to attach.
Solidity calls its own security recommendations incomplete, and ethereum dot org warns that audits do not catch every bug. That is why every row asks for evidence, not a tick.
The checks follow five attack surfaces: access control, external calls, math and accounting, oracles and integrations, then upgrades and deployment.
Log each check with its evidence and a status. A pass needs a second reviewer. A fail is an issue with an owner. Not tested goes to auditors, and not applicable needs a written reason.
In Guardian's Clanker review, extension authority stayed set during an ETH refund, so a creator-controlled recipient could reenter. That High finding is resolved, and check twenty three is the row that targets it.
In Guardian's Papertrade review, each division rounded down, yet summed reward obligations could exceed the reserve and block a claim. Checks thirty eight and forty two catch this with invariant tests.
Four gaps teams often miss. Check oracle freshness with updatedAt. Safe ERC twenty handles failure signals, not fees or rebases. Lock implementation initializers. And pin a released compiler, since nightly builds have no known bugs record.
Read the full article for all seventy five rows and the evidence each one needs. Then bring rows still marked fail or not tested to Guardian's smart contract audit, with two independent teams and fix review.
How to Use This Audit Checklist
Solidity's own security page calls its recommendations incomplete, and ethereum.org warns that audits do not catch every bug. So each row asks for evidence, not a tick.[1, 2]
Launch has its own pre-mainnet security checklist; logistics are in how to prepare for a smart contract audit.
Copy each check into your tracker under its own ID, with function, property, evidence, reviewer and status. The category is the finding class a failure would fall under.
- Pass: evidence attached and checked by a second reviewer.
- Fail: an open issue with an owner.
- Not tested: flagged to auditors as an open question.
- N/A: a written reason, such as no signatures in scope.
Access Control and Privileged Functions
Bug patterns behind these rows are in access control vulnerabilities in smart contracts.
In OpenZeppelin's AccessControl, a role's admin can grant and revoke it, so review admins too.[3]
Ownable2Step makes the new owner accept the transfer. Renouncing ownership leaves every onlyOwner function unusable.
ERC-2612 permits need a valid signature, the current nonce, a nonzero owner and an unexpired deadline. Success increments the nonce.[4]
| ID | Check | Category | Evidence |
|---|---|---|---|
| 01 | Inventory public, external and inherited entry points | Access Control | Entry-point list |
| 02, 03 | Give each entry point a caller policy; test refusals | Access Control | Permission matrix; negative tests |
| 04 | Bind account and position IDs to their caller | Access Control | Foreign-ID tests |
| 05, 06 | Map each role's admin; test grant, revoke and rotation | Access Control | Role-admin map; lifecycle tests |
| 07 | Check ownership acceptance and what renouncing disables | Access Control | Transfer and renounce tests |
| 08 | Review timelock proposers, executors and any bypass | Access Control | Timelock role list |
| 09 | Bound privileged fee, cap and recipient changes | Validation | Tests at each limit |
| 10 | Verify pause and unpause permissions separately | Access Control | Pause role tests |
| 11 | Bind signatures to every authorized field | Signature Verification | Typed-data spec |
| 12, 13 | Consume nonces once; separate domains by contract and chain | Replay | Replay and cross-chain tests |
| 14 | Reject expired, invalid and zero-owner authorizations | Signature Verification | Negative signature tests |
| 15 | Never authorize with tx.origin | Access Control | Code search |
External Calls, Reentrancy and Callbacks
An external call can hand control away before your function finishes; see external call vulnerabilities in Solidity. Solidity's guidance says a called contract can change the state of other contracts you depend on.[1]
Guardian's Clanker review shows why check 23 matters (Main Review H-06, High, resolved). Extension authority stayed set during an ETH refund, so a creator-controlled recipient could reenter and drain protocol fees the core held.[5]
OpenZeppelin's transient guard needs EIP-1153 support. Solidity advises clearing transient locks when each call ends.[6, 7]
| ID | Check | Category | Evidence |
|---|---|---|---|
| 16 | List every external call, including token hooks | Reentrancy | Call inventory |
| 17 | Update state before handing over control | Reentrancy | Call-order review |
| 18, 19 | Protect same-function, inherited and alternate entry points | Reentrancy | Reentry tests |
| 20, 21 | Review shared state and views read mid-call | Reentrancy | Cross-contract state map |
| 22 | Authenticate callback senders and expected context | Access Control | Direct-call tests |
| 23 | Clear temporary authority before refunds and failure callbacks | Reentrancy | Exceptional-path review |
| 24 | Keep guards from blocking legitimate composition | Reentrancy | Integration tests |
| 25 | Confirm transient-storage support and lock cleanup | Configuration | Chain list; same-transaction tests |
| 26 | Restrict delegated execution targets | Access Control | Allowed-target list |
| 27, 28 | Handle low-level call failure; validate and bound return data | Validation | Failure and large-return tests |
| 29 | Bound loops and batch work | DoS | Maximum-size gas tests |
| 30 | Stop one recipient's failure blocking unrelated payouts | DoS | Reverting-recipient test |
Math, Rounding and Accounting
Solidity integer division truncates toward zero. Division by zero still reverts inside unchecked blocks.[8]
Shifts are not overflow-checked, and checked arithmetic does not make narrowing conversions safe.
ERC-4626 requires its conversion functions to round down. Each preview must be no more favorable to the user than the matching operation in the same transaction.[9]
Guardian's Papertrade review shows what checks 38 and 42 catch. Each division rounded down, yet summed reward obligations could exceed the reserve and block a claim (M-03, Medium, resolved).[10]
Foundry invariant tests with ghost variables suit checks 38 to 45.[11]
| ID | Check | Category | Evidence |
|---|---|---|---|
| 31 | Document units and decimal scales | Validation | Units table |
| 32 | Set a rounding direction for every operation | Rounding | Rounding spec |
| 33 | Handle every zero denominator | Arithmetic | Zero-input tests |
| 34 | Bound intermediate results, not only outputs | Arithmetic | Maximum-value tests |
| 35 | Justify every unchecked block | Arithmetic | Comment per block |
| 36, 39 | Check narrowing casts, shifts, signed boundaries and truncation | Arithmetic | Boundary tests |
| 37 | Set minimum amounts; handle zero-result conversions | Rounding | Dust-amount tests |
| 38 | Compare split operations with one aggregate operation | Rounding | Split-versus-whole fuzz test |
| 40 | Reconcile assets, liabilities and reserved funds | Accounting | Solvency invariant |
| 41, 42 | Charge each fee once; keep reward obligations within funding | Accounting | Fee and reward invariants |
| 43 | Keep parallel records consistent | Logical Error | Sync invariant |
| 44, 45 | Cover empty supply, first and last users, donations, deficits and final exits | Accounting, DoS | Edge-state tests |
Oracles, Tokens and Integrations
Chainlink documents updatedAt for freshness and marks answeredInRound deprecated. Heartbeat updates can lag, so set limits per application.[12, 13]
Chainlink's sequencer uptime example checks availability and a grace period but is not production code as supplied.[14]
ERC-20 callers must handle a false return, and decimals is optional. SafeERC20 handles failure signals, not fees or rebases.[15, 16]
Uniswap's v3 docs say fee-on-transfer tokens do not work with its routers. Other quirks are in our weird ERC-20 tokens checklist.[17]
OpenZeppelin warns against safeIncreaseAllowance and safeDecreaseAllowance while an ERC-7674 temporary allowance is nonzero.
ERC-4626 warns that vault previews may not suit price oracles.[9]
| ID | Check | Category | Evidence |
|---|---|---|---|
| 46 | Verify dependency addresses on every network | Configuration | Address table per chain |
| 47 | Verify price denomination and decimal normalization | Validation | Normalization tests |
| 48, 49 | Enforce freshness and acceptable answer ranges per feed | Oracle | Stale-price tests |
| 50 | Define behavior when price data is unavailable | DoS | Failure-mode spec |
| 51 | Handle sequencer outages and recovery where applicable | Oracle | Outage tests |
| 52 | Check whether a price input can move without a real change in value | Oracle | Manipulation cost note |
| 53 | Handle false and missing token return values | Integration | Mock-token tests |
| 54 | Support or exclude transfer-fee and rebasing tokens explicitly | Accounting | Token allowlist or tests |
| 55, 56 | Model callback tokens; document dependency pause and upgrade powers | Reentrancy, Integration | Dependency register |
| 57, 58 | Review both allowance types; let permit fail, then check the real allowance | Access Control, Integration | Allowance and permit tests |
| 59 | Distinguish vault conversions, previews and limits | Integration | Vault integration tests |
| 60 | Enforce intended amounts, recipients and deadlines | Validation | Slippage and deadline tests |
Upgrades, Deployment and Configuration
Solidity publishes known bugs per compiler version and warns that an incompatible EVM target can produce incorrect or failing behavior.[18, 19]
OpenZeppelin recommends _disableInitializers in implementation constructors; see initializer vulnerabilities. Its plugins flag missing or repeated parent initializers but do not validate reinitializers by default.[20]
Current ERC1967Proxy rejects empty initialization data unless overridden, and UUPS implementations must restrict _authorizeUpgrade.[21]
| ID | Check | Category | Evidence |
|---|---|---|---|
| 61 | Freeze the review commit and a scope manifest | Configuration | Commit and file list |
| 62 | Pin dependencies, optimizer and build settings | Configuration | Lockfile and build config |
| 63 | Check known bugs for your exact compiler version | Compiler | Known-bugs review note |
| 64 | Match the EVM target to each chain | Configuration | Per-chain target table |
| 65 | Resolve or justify every compiler warning | Compiler | Clean build log |
| 66, 67 | Guard initializers; run each parent initializer exactly once | Access Control, Logical Error | Double-initialization tests |
| 68 | Lock implementation initializers | Access Control | Constructor review |
| 69 | Initialize each proxy in its deployment transaction | Access Control | Deployment script review |
| 70 | Restrict every upgrade route | Access Control | Unauthorized-upgrade tests |
| 71, 72 | Validate storage layout; review reinitializers and migrations | Upgradeability | Layout diff and migration tests |
| 73, 74 | Assert script parameters, final roles, artifacts and proxy implementation | Configuration | Script assertions; bytecode match |
| 75 | Rerun all tests after every fix | Validation | CI run per fix |
Take the open rows to an independent review
Guardian's smart contract audit puts two independent teams on one scope, with invariant fuzzing, human-verified findings and fix review. Bring the rows still marked Fail or Not tested.
Frequently Asked Questions
Can we run the compiler checks on a nightly Solidity build?
Not fully. Solidity's known-bugs list does not track unreleased or nightly versions, so check 63 has no record to compare against. Pin a released version for the build you send to review.[18]
Is a Chainlink sequencer uptime feed available on every L2?
No. Existing supported uptime feeds continue, but Chainlink says it has stopped expanding them to additional networks. On other chains, document how the protocol detects an outage, or mark check 51 N/A with a reason.[14]
Are the checks ranked by severity?
No. IDs follow attack surface, not risk. A failed row's rating depends on impact and likelihood in your system, as smart contract audit severity levels explains.
Related reading
- sneaky solidity bugs
- erc4626 security checklist
- ways to miss vulnerabilities in an audit
- solidity security best practices
Sources
- Solidity: Security Considerations
- ethereum.org: Smart Contract Security
- OpenZeppelin: Access Control
- Ethereum Improvement Proposals: ERC-2612: Permit Extension for EIP-20 Signed Approvals
- Guardian: Clanker Protocol Review: H-06
- OpenZeppelin: Utils
- Solidity: Contracts
- Solidity: Types
- Ethereum Improvement Proposals: ERC-4626: Tokenized Vaults
- Guardian: Papertrade Protocol: M-03
- Foundry: Invariant Testing
- Chainlink: Data Feeds API Reference
- Chainlink: What are Chainlink Data Feeds?
- Chainlink: L2 Sequencer Uptime Feeds
- Ethereum Improvement Proposals: ERC-20: Token Standard
- OpenZeppelin: ERC20
- Uniswap: Token Integration Issues: official documentation source
- Solidity: List of Known Bugs
- Solidity: Using the Compiler
- OpenZeppelin: Writing Upgradeable Contracts
- OpenZeppelin: Proxy
- OpenZeppelin: ERC-4626
- Guardian: LayerZero Console EVM Updates: I-02
- Guardian: Smart Contract Audit Services