agentleFS
iktok90-designiktok90-design/ai-smart-contract-auditor/skills/stylus-rust/SKILL.md

stylus-rust

Skill36 starsChanged 5 months agoMITView on GitHub

Detect bug classes specific to Arbitrum Stylus (Rust→WASM) contracts — storage aliasing & EVM state-cache coherence, msg::value / #[payable] handling, external-call reentrancy, panic-on-attacker-input DoS, release-mode integer wrapping, host-IO (evm::) misuse, #[entrypoint]/#[public] access control, and lossy U256 conversions. Activate on any `.rs` file with `use stylus_sdk`, `#[entrypoint]`, `#[storage]`, `sol_storage!`, `#[public]`, or `#[payable]`.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Signal: every method in a #[public] impl is externally callable.

    line 74

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Stylus (Rust) detection

When this applies

  • Any .rs file importing stylus_sdk (use stylus_sdk::{...})
  • Macros #[entrypoint], #[storage], sol_storage!, #[public], #[payable]
  • Host-IO calls: evm::, msg::, block::, contract::, call:: (Call::new, transfer_eth, RawCall)
  • Stylus types StorageU256, StorageMap, StorageVec, alloy_primitives::U256/Address
  • A Cargo.toml declaring stylus-sdk with crate-type = ["lib", "cdylib"]

Stylus runs the SAME EVM state and shares the SAME external-call surface as Solidity. Rust safety does NOT remove EVM-level footguns — it adds new ones (panics, wrapping arithmetic, aliasing).

Detection patterns

Storage aliasing / stale local copy of EVM state (HIGH)

let mut bal = self.balances.get(from);      // ← copies value out of storage
do_external_call();                          // callee may mutate self.balances
self.balances.insert(from, bal - amount);    // ← writes back STALE value

Signal: a .get() cached in a local, an intervening call/host-IO, then a .set()/.insert() of the stale local. Stylus storage reads are snapshots, not live references — re-read after any external call.

Reentrancy via external call before state finalization (CRITICAL)

pub fn withdraw(&mut self) -> Result<(), Vec<u8>> {
    let amt = self.balance.get(msg::sender());
    call::transfer_eth(msg::sender(), amt)?;     // ← control leaves contract
    self.balance.setter(msg::sender()).set(U256::ZERO);  // ← AFTER the call
    Ok(())
}

Signal: transfer_eth / Call::new().call(...) / RawCall before the storage write that zeroes the credited amount. Reentrancy is OFF by default but Call::new() and transfer_eth still re-enter; CEI still required.

Panic / unwrap on attacker input = revert-DoS (HIGH)

let idx: usize = msg::data()[0].into();
let entry = self.items.get(idx).unwrap();    // ← panics → reverts whole tx

Signal: .unwrap(), .expect(), indexing [i], or arithmetic that can panic on caller-controlled input. A Rust panic traps the WASM and reverts — fine for the victim caller, but if it's on a path others depend on (batch settlement, queue processing) it's a griefing DoS. Return Result/Err, never panic on untrusted input.

Release-mode integer wrapping (HIGH)

let total = a + b;          // ← in --release this WRAPS, no panic, no revert
self.supply.set(total);

Signal: bare +/-/* on u64/u128/U256 in contract math. Stylus ships in release mode where overflow checks are OFF (wraps silently). alloy U256 also wraps on +. Use checked_add/checked_sub/checked_mul (or overflow-checks = true in Cargo.toml).

#[payable] vs unguarded msg::value (HIGH)

#[public]
impl Token {
    pub fn deposit(&mut self) {               // ← NOT #[payable]
        let v = msg::value();                  // value forced to 0 by ABI? no —
        self.credit.setter(msg::sender()).set(v);
    }
}

Signal: reading msg::value() in a non-#[payable] method (sent ETH gets stuck / call reverts depending on path), OR a #[payable] method that credits msg::value() without bounds, OR forgetting msg::value() is per-call (re-reads of it don't re-validate).

#[public] / #[entrypoint] missing access control (HIGH)

#[public]
impl Vault {
    pub fn set_owner(&mut self, who: Address) {   // ← anyone can call
        self.owner.set(who);
    }
}

Signal: every method in a #[public] impl is externally callable. Admin/set_*/mint/upgrade methods with no if msg::sender() != self.owner.get() { return Err(...) } check. There is no implicit visibility gate — #[public] == external.

Lossy / truncating U256 conversions (MEDIUM-HIGH)

let n: u64 = amount.to::<u64>();        // ← truncates if amount > u64::MAX
let small: u128 = big_u256.wrapping_to();

Signal: .to::<u64>(), as u64, wrapping_to, or try_into().unwrap() from U256 to a narrow type used in accounting. Use U256::try_into and handle the error.

Host-IO host call assumptions (MEDIUM)

let ok = RawCall::new().call(target, &data)?;  // ← gas, return-size, success

Signal: RawCall/Call results where success bool and return bytes aren't both checked, or hard-coded gas via .gas(...) that strands recipients. Same low-level-call hygiene as Solidity ([[unchecked-calls]]).

Severity rubric

PatternSeverityNotes
External call before state finalize (CEI broken)CriticalDirect fund drain, same as Solidity
Stale storage cache across external callHighAliasing reads are snapshots
Release-mode wrapping in supply/balance mathHighOverflow checks off by default
unwrap/index/panic on attacker inputHighRevert-DoS on shared paths
msg::value mishandled vs #[payable]HighStuck or unbounded ETH
#[public] admin method, no sender checkHighAnyone calls privileged fn
Truncating U256→u64/u128 in accountingMediumValue-dependent
Unchecked RawCall/hardcoded gasMediumRecipient-dependent

Remediation patterns

  1. CEI ordering — write all storage effects before transfer_eth/Call. Re-read storage after any external call instead of trusting a cached local.
  2. checked_* arithmetic everywhere, or set overflow-checks = true under [profile.release] in Cargo.toml (accept the WASM size/gas cost).
  3. Never unwrap/expect/raw-index on untrusted input — propagate Result<_, Vec<u8>> and return Err.
  4. Gate privileged #[public] methods with an explicit msg::sender() owner/role check; consider a reentrancy guard storage bool for callback paths.
  5. Make value-receiving methods #[payable] and validate msg::value() bounds; reject unexpected value in non-payable methods.
  6. Use try_into() for U256 narrowing and handle the error rather than as/wrapping_to.

False-positive notes

  • checked_add(...).ok_or(...)? and similar are already safe — don't flag wrapping there.
  • A #[public] method that is a pure view (&self) with no state writes and no value isn't an access-control issue.
  • .unwrap() on a value the contract itself just constructed (not attacker-derived) is generally fine — focus on untrusted-input paths.
  • Stylus reentrancy is disabled unless the reentrant feature is enabled; still flag CEI violations since the feature flag is set per-project and callbacks can re-enter cross-contract.

Related

  • [[reentrancy]] — Stylus shares the EVM external-call reentrancy surface
  • [[integer-issues]] — release-mode wrapping + lossy U256 conversions
  • [[access-control]] — #[public]/#[entrypoint] has no implicit visibility gate

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.