Review Sui Contracts

AI-assisted review of a developer's Sui Move code against OpenZeppelin Contracts for Sui: does it use the library's on-chain primitives correctly, and where does it deviate from the library's examples/, doc-comments, and documented conventions?

This is an AI code review, not a formal security audit. It supports the audit process by flagging misuse and deviations early; it does not replace an independent audit, and a clean review is not an assurance of safety.

For building an integration use develop-secure-contracts; for scaffolding a project use setup-sui-contracts. This skill reviews code that already exists.

Establish the reference before reviewing

You cannot flag "incorrect use" or "deviation" without the correct pattern in hand — so discover it from the library's own metadata, never from memory, using the discovery chain the develop-secure-contracts skill defines. Read the primitive's API and behavior at the revision the integrator builds against — resolve each OZ dependency's pinned rev from Move.lock and use that source / those doc-comments, not main. Read the library-wide docs — llms.txt, the catalogs, ARCHITECTURE.md, STYLEGUIDE.md — from the repo's current main instead: they evolve continuously and often do not exist at an older pinned rev. For each OZ primitive the code uses, load its authoritative pattern from:

• the package's examples/ if it has them, otherwise the module-header doc-comment's idiomatic-usage block — the canonical recipe (some packages ship only the latter);
• the modules' doc-comments and generated API reference (reached via the catalog's Docs link);
• ARCHITECTURE.md (capability model, object ownership, initialization) and STYLEGUIDE.md (conventions).

Review the code against those, not against assumptions.

What to review

• Correct use of OZ primitives. Does the code thread the capability/witness the component defines (rather than hand-rolling sender checks or re-implementing what the library already provides), call functions with the right arguments, and handle the documented abort conditions?
• Who chooses the type argument — and is it the right one. For each OZ primitive parameterized by a witness or generic/phantom type, check who supplies it and whether it is the correct authority. A caller-chosen type is an authority the caller controls, so the trusted party must fix it, not the caller. And a developer-fixed type must be the right one for the action — a privileged operation gated by a lower-privilege role's Auth (or any weaker witness than the guarantee needs) is a privilege-escalation bug.
• Deviations from the reference pattern. Compare the flow against the closest examples/ recipe (or the doc-comment's idiomatic-usage block) and flag divergences that weaken the guarantees it preserves — e.g. an object shared vs. embedded against its intended ownership model, a required setup step skipped, a returned value or receipt ignored, or an invariant the doc-comment states being bypassed.
• Copied library source. Flag any OZ module source copied into the project instead of imported via MVR — copies miss security updates.
• Invariants upheld. Check that the code preserves the invariants the OZ components document (in their doc-comments and ARCHITECTURE.md) — e.g. an object that must stay shared, a value that must remain bounded, a capability that must not be exposed. Flag any path that can break one.
• Test coverage of the integration. Check that the tests exercise the abort and branch paths the OZ usage introduces (role-denied, expired, over-limit, wrong-type, and similar), not just the happy path — untested failure paths are where integration bugs hide.
• Known issue classes (heuristic). Check the code for the classes of issue that recur in OZ-on-Sui / Move code — capability handling, rounding, object ownership, initialization — and raise suspicion where the code touches them. (The library's published audits, under audits/, inform these classes; the reports themselves are PDFs, not a findings feed.) A suspicion-raising self-check, not a reproduction of the audits or a security audit.

Code quality — enforce the conventions

Treat OpenZeppelin's Sui conventions as a first-class review dimension: report violations as findings, not optional suggestions. The rules and the review procedure are the single source of truth — apply them, don't restate them:

• STYLEGUIDE.md — the codified rules (naming, section ordering, imports, idiomatic Move 2024, testing, docs, lint suppression).
• .claude/commands/code-quality.md — the canonical procedure that checks .move files against STYLEGUIDE.md; run it against the code under review.
• Lint is not optional. Build and test with the project's --build-env and --lint (see setup-sui-contracts for the exact invocation); per the styleguide, warnings must be fixed, never suppressed with #[allow(...)]. --lint enforces the compiler-checkable subset independently, so if the STYLEGUIDE.md fetch is momentarily unreachable, retry rather than skipping conventions.

Report each violation against the STYLEGUIDE.md section it breaks.

Verify before reporting

The checks above optimize for catching issues, so before writing anything, try to disprove each finding: quote the exact line and confirm the problem is really there; confirm it is a genuine deviation and not an intentional, documented choice; and confirm you can state a concrete one-line fix. Drop whatever does not survive all three, and keep the severity bar high — a short review a developer trusts beats a long one they skim.

Output

Produce one review, findings ordered by severity. Each finding: file:line — what's wrong (one clause) → the fix (one clause) — and the OZ example or doc-comment it deviates from, so the developer can compare directly. A short preamble is useful — the scope/boundary call and the build/test/lint status — and on a clean integration a single "verified and cleared" line is fine; keep the body to findings, not a recap.

If the project builds, run its checks first (sui move build / test, per setup-sui-contracts) so the review reflects compiling code, and report build/test failures separately from the pattern findings.

Boundaries

• Reviews the integrator's code — how it uses the library — not the OpenZeppelin library internals themselves.
• Not a security audit — no exhaustive threat modeling or completeness guarantee; recommend a formal audit for production code.
• Complements develop-secure-contracts (integration) and setup-sui-contracts (scaffolding); it neither builds nor scaffolds.