smart-contract-review
Skill0 starsChanged 3 months agoMITView on GitHub
Review a Solidity smart contract for the vulnerabilities that actually cause fund loss, before it is deployed. Use when the user shares a .sol file or contract address, asks to audit a contract, review DeFi code, or check whether a smart contract is safe.
In this repository11 other files it gives its agents.
- agent-verificationSkill
- claim-evidenceSkill
- crypto-news-digestSkill
- defi-risk-assessmentSkill
- evidence-firstSkill
- flaky-test-detectorSkill
- on-chain-analystSkill
- prompt-injection-hardeningSkill
- refactor-verifierSkill
- skill-security-auditSkill
- test-integrity-guardSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 19
Do not review a contract you have not read end to end, including inherited contracts and interfaces it calls.
- Very specific: line 31
Look for the pattern: external call → state change.
- Distinctive: line 72
Check whether a malicious actor could profitably distort the input the contract reads.
- Distinctive: line 99
Every finding must point at the exact function/line and explain the exploitation path, not just name a category.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Smart Contract Review
Smart-contract bugs are permanent and usually expensive: code on-chain cannot be patched after the fact, and a missed vulnerability means real funds lost. This skill is a focused security review for Solidity, ordered by what has historically caused the most real-world loss, not by textbook completeness.
Read the contract first, fully
Do not review a contract you have not read end to end, including inherited
contracts and interfaces it calls. Most misses come from reviewing the entry
point while skipping a base contract, a library, or an external call target.
The vulnerabilities that lose money, in order
1. Reentrancy
A contract that calls an external contract, then updates its own state
afterward, can be re-entered before the state settles — the classic
withdraw that sends funds before zeroing the balance.
- Look for the pattern: external call → state change. The state change must
come first, or the call must be
nonReentrant(OpenZeppelinReentrancyGuard). - Pay special attention to ERC-777/ERC-721 callback paths and native
call{value:}.
2. Access control
Who can call the state-changing functions? Check every onlyOwner /
onlyRole / custom modifier:
- A missing modifier on a
setX/withdraw/mint/upgradefunction. - An over-permissive role (
DEFAULT_ADMIN_ROLEhanded to the wrong address). tx.originused for authorization.tx.originis the externally-owned account that started the whole transaction chain, notmsg.sender(the immediate caller) — they're equal for a direct call, but diverge the moment any contract is interposed. An attacker can trick the real owner into calling a malicious contract, which then calls the target on the owner's behalf:msg.senderis the malicious contract (correctly rejected), buttx.originis still the legitimate owner (wrongly authorized). Usemsg.senderfor authorization checks.initialize()left callable (an uninitialized proxy can be taken over).
3. Integer overflow / underflow
Solidity ≥0.8.0 reverts on overflow by default, but:
unchecked { }blocks bypass it — audit every one.- Older pragmas (
^0.7and below) have no built-in check at all. - Casting and bit-shift edge cases can still wrap.
4. Unchecked external return values and failed sends
transfer/sendare capped at 2300 gas and can silently fail on contracts with a receive/fallback that costs more.call(the low-level one) returns a bool — an ignoredboolmeans a failed call is treated as success.
5. Oracle / price manipulation
- A price derived from a spot DEX pair (
getReserves()) is flash-loan manipulable — anyone can move the price within one transaction. - Check whether a malicious actor could profitably distort the input the contract reads.
6. Flash-loan and governance attacks
- Functions that let a large, instant loan change a parameter the contract trusts (governance voting on spot balances, minting based on current TVL).
delegatecallto an untrusted/upgradeable address — full code-execution risk.
7. Front-running and MEV
- Transactions where the outcome depends on who gets mined first (sandwichable swaps, reveals after commit that can be front-run).
Output format
Contract: <name / address>
Severity findings:
- CRITICAL: <finding> — <location> — <why it loses funds> — <fix>
- HIGH: ...
- MEDIUM: ...
- LOW / informational: ...
Verdict: DO NOT DEPLOY | FIX BEFORE DEPLOY | LGTM (with caveats)Every finding must point at the exact function/line and explain the exploitation path, not just name a category. A category without an exploitation path is a guess, not a finding.
Boundaries
This is a review, not an audit certificate. State clearly that a clean review is not a guarantee, recommend a professional audit + testnet/formal verification for anything holding real value, and do not claim to have found "all" bugs — only the ones you can demonstrate.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

