Review Verification Protocol

This protocol MUST be followed before reporting any code review finding. Skipping these steps leads to false positives that waste developer time and erode trust in reviews.

Anti-confabulation (gate 0 — runs before every other gate)

Before issuing any verdict — flag, reject, or downgrade a finding — you MUST echo the exact artifact you are judging, quoted from a source you read in this turn:

• For a code finding: the file:line plus the cited code, read freshly now (not recalled from earlier in the session).
• For a diff review: the actual diff hunk under review.

The artifact is the only source of truth. Never infer what you are reviewing from the branch name, the working directory, surrounding files, or recollection. If your mental model differs from the freshly read source, the source wins. A verdict issued without a same-turn echo of its target is invalid — emit the echo first, or do not emit the verdict.

This gate exists because an LLM under contextual priming will confidently flag code that is not in the file. It runs before the hard gates below.

Hard gates (sequenced)

Complete these in order before you add a finding. Skip a gate only when it clearly does not apply (e.g. skip the usages gate if the finding is not about dead code or “unused”).

1. Read scope — Pass: You name the exact file path(s) and the function, impl, or macro_rules! block you read in full (not only a diff hunk or partial snippet).
2. Usages (dead / unused) — Pass: You ran a repo-wide reference search (rg, IDE references, or equivalent) and either state zero matches for the symbol you call unused, or list each match and why it still supports the finding.
3. Surrounding behavior — Pass: You checked callers, trait impls, #[cfg], or error propagation that could make the pattern intentional; note one concrete checked location (path + rough location) or state “none relevant after search.”
4. Edition and API — Pass: You opened the relevant Cargo.toml for the crate under review and either quote the [package] edition = "..." line or state the default edition applies and name the manifest path you checked.
5. Wrong vs style — Pass: In one sentence, you explain why the code is incorrect, unsound, or risky for this project—not merely a different valid style.

Pre-Report Verification Checklist

Before flagging ANY issue, verify:

• [ ] I read the actual code - Not just the diff context, but the full function/impl block
• [ ] I searched for usages - Before claiming "unused", searched all references
• [ ] I checked surrounding code - The issue may be handled elsewhere (trait impls, error propagation)
• [ ] I verified syntax against current docs - Rust edition, crate versions, and API changes
• [ ] I checked the project's Rust edition - Edition 2021 vs 2024 changes what is required vs optional (see Edition-Aware Review)
• [ ] I distinguished "wrong" from "different style" - Both approaches may be valid
• [ ] I considered intentional design - Checked comments, project conventions (e.g. AGENTS.md or CLAUDE.md), architectural context

Verification by Issue Type

"Unused Variable/Function"

Before flagging, you MUST:

1. Search for ALL references in the codebase (grep/find)
2. Check if it's pub and used by other crates in the workspace
3. Check if it's used via derive macros, trait implementations, or conditional compilation (#[cfg])
4. Verify it's not a trait method required by the trait definition

Common false positives:

• Trait implementations where the method is defined by the trait
• #[cfg(test)] items only used in test builds
• Derive-generated code that uses struct fields
• Types used via From/Into conversions

"Missing Error Handling"

Before flagging, you MUST:

1. Check if the error is handled at a higher level (caller propagates with ?)
2. Check if the crate has a top-level error type that wraps this error
3. Verify the unwrap() isn't in test code or after a safety-ensuring check

Common false positives:

• unwrap() in tests and examples (expected pattern)
• expect("reason") after validation (e.g., regex::Regex::new on a literal)
• Error propagation via ? (the caller handles it)
• let _ = tx.send(...) — intentional when receiver may have dropped

"Unnecessary Lifetime" / RPIT Capture (Edition 2024)

Before flagging, you MUST:

1. Check the project's Rust edition in Cargo.toml
2. In edition 2024, -> impl Trait captures ALL in-scope lifetimes by default
3. A lifetime that appears "unnecessary" may be implicitly captured — the code is correct
4. If the author uses + use<'a> syntax, this is precise capture control, not a mistake

Common false positives:

• Lifetime parameters on functions returning impl Trait — edition 2024 captures them implicitly
• + use<'a, T> syntax — this is the new precise capturing syntax, not an error
• Removing an explicit lifetime bound that edition 2024 now provides automatically

"Missing Unsafe Block" (Edition 2024)

Before flagging, you MUST:

1. Check if the code is inside an unsafe fn
2. In edition 2024, unsafe_op_in_unsafe_fn is deny-by-default — unsafe operations inside unsafe fn REQUIRE explicit unsafe {} blocks
3. This is edition-required behavior, not unnecessary verbosity

Common false positives:

• unsafe {} blocks inside unsafe fn — REQUIRED in edition 2024, not redundant
• unsafe extern "C" {} — REQUIRED in edition 2024, not optional
• #[unsafe(no_mangle)] / #[unsafe(export_name)] — REQUIRED in edition 2024

"Unnecessary Clone"

Before flagging, you MUST:

1. Confirm the clone is actually avoidable (borrow checker may require it)
2. Check if the value needs to be moved into a closure/thread/task
3. Verify the type isn't Copy (clone on Copy types is a no-op)
4. Check if the clone is in a hot path (test/setup code cloning is fine)

Common false positives:

• Arc::clone(&arc) — this is the recommended explicit clone for Arc
• Clone before tokio::spawn — required for 'static bound
• Clone in test setup — clarity over performance

"Potential Race Condition"

Before flagging, you MUST:

1. Verify the data is actually shared across threads/tasks
2. Check if Mutex, RwLock, or atomic operations protect the access
3. Confirm the type doesn't already guarantee thread safety (e.g., Arc<Mutex<T>>)
4. Check if the "race" is actually benign (e.g., logging, metrics)

Common false positives:

• Arc<Mutex<T>> — already thread-safe
• Tokio channel operations — inherently synchronized
• std::sync::atomic operations — designed for concurrent access

"Performance Issue"

Before flagging, you MUST:

1. Confirm the code runs frequently enough to matter
2. Verify the optimization would have measurable impact
3. Check if the compiler already optimizes this (iterator fusion, inlining)

Do NOT flag:

• Allocations in startup/initialization code
• String formatting in error paths
• Clone in test code
• .collect() on small iterators

Severity Calibration

Critical (Block Merge)

ONLY use for:

• unsafe code with unsound invariants
• SQL injection via string interpolation
• Use-after-free or memory safety violations
• Data races (concurrent mutation without synchronization)
• Panics in production code paths on user input

Major (Should Fix)

Use for:

• Missing error context across module boundaries
• Blocking operations in async runtime
• Mutex guards held across await points
• Missing transaction for multi-statement database writes

Minor (Consider Fixing)

Use for:

• Missing doc comments on public items
• String parameters where &str would work
• Suboptimal iterator patterns
• Missing #[must_use] on functions with important return values

Informational (No Action Required)

Use for:

• Suggestions for newtypes, builder patterns, or type state
• Performance optimizations without measured impact
• Suggestions to add #[non_exhaustive]
• Refactoring ideas for trait design

These are NOT review blockers.

Do NOT Flag At All

• Style preferences where both approaches are valid (e.g., if let vs match for single variant)
• Optimizations with no measurable benefit
• Test code not meeting production standards
• Generated code or macro output
• Clippy lints that the project has intentionally suppressed

Valid Patterns (Do NOT Flag)

Rust

Async/Tokio

Testing

General

Context-Sensitive Rules

Ownership

Flag unnecessary .clone() ONLY IF:

• [ ] In a hot path (not test/setup code)
• [ ] A borrow or reference would work
• [ ] The clone is not required for Send/'static bounds
• [ ] The type is not Copy

Error Handling

Flag missing error context ONLY IF:

• [ ] Error crosses a module boundary
• [ ] The error type doesn't already carry context (thiserror messages)
• [ ] Not in test code
• [ ] The bare ? loses meaningful information about what operation failed

Unsafe Code

Flag unsafe ONLY IF:

• [ ] Safety comment is missing or doesn't explain the invariant
• [ ] The unsafe block is broader than necessary
• [ ] The invariant is not actually upheld by surrounding code
• [ ] A safe alternative exists with equivalent performance

Edition 2024 unsafe changes — check Cargo.toml edition before flagging:

• unsafe {} inside unsafe fn is required (not style) in edition 2024
• unsafe extern "C" {} is required in edition 2024 — bare extern "C" {} is a compile error
• #[unsafe(no_mangle)] and #[unsafe(export_name)] are required in edition 2024
• In edition 2021, these patterns are optional style choices — do not require them

Edition-Aware Review

BEFORE flagging any edition-specific pattern, check Cargo.toml for the project's edition:

[package]
edition = "2024"  # or "2021", "2018"

Edition 2024 changes that affect review findings:

If edition is not specified, Rust defaults to edition 2015. Most modern projects use 2021 or later.

Cross-reference: The rust-code-review and rust-best-practices skills provide edition-specific code review guidance and idiomatic patterns.

Macro-Specific Verification

"Macro Hygiene Issue"

Before flagging, you MUST:

1. Verify the identifier actually leaks — types, modules, and functions are NOT hygienic in macro_rules!
2. Check if $crate is used correctly for exported macros (not crate or self)
3. Confirm ::core:: / ::alloc:: paths are needed (only for macros used in no_std contexts)
4. Check whether the macro is internal-only or #[macro_export]

Common false positives:

• Non-hygienic type names in internal macros — only matters for exported macros
• $crate not used in macros that are only pub(crate) — $crate is for cross-crate usage
• Using ::std:: in macros for std-only crates — only flag if crate supports no_std

"Procedural Macro Performance"

Before flagging, you MUST:

1. Verify the macro is actually in a proc-macro crate (check Cargo.toml for proc-macro = true)
2. Check if syn features are minimized (full syn with "full" feature vs selective features)
3. Confirm compile-time impact is meaningful (proc macros used across many files vs one-off)

"Wrong Fragment Type"

Before flagging, you MUST:

1. Verify the suggested fragment type actually works in that position
2. Check if :tt is intentionally used for flexibility (common in TT munching patterns)
3. Confirm :expr greediness issues actually manifest (test with the macro's actual call sites)

FFI-Specific Verification

"Missing repr(C)"

Before flagging, you MUST:

1. Confirm the type actually crosses the FFI boundary (passed to/from C code)
2. Check if the type is only used on the Rust side of the FFI wrapper
3. Verify there isn't a #[repr(transparent)] wrapper instead

Common false positives:

• Internal Rust types that are converted before FFI call — only the FFI-facing type needs repr(C)
• Types used with repr(transparent) newtype wrappers — the wrapper handles layout
• Opaque pointer types (*mut c_void) — no layout guarantee needed

"FFI Safety"

Before flagging, you MUST:

1. Check if the unsafe FFI call has a SAFETY comment documenting invariants
2. Verify ownership transfer is actually ambiguous (check for Box::into_raw/Box::from_raw pairs)
3. Confirm CString lifetime issues are real (the CString must outlive the pointer passed to C)
4. Check if callback unwinding is actually possible (pure data functions can't panic across FFI)

Common false positives:

• extern "C" fn callbacks that never panic — catch_unwind not needed
• *const c_char from CStr::as_ptr() held within the same scope — lifetime is fine
• Bindgen-generated code with unsafe — bindgen output is inherently unsafe-heavy by design

Concurrency-Specific Verification

"Memory Ordering Too Weak"

Before flagging, you MUST:

1. Verify the atomic is actually shared between threads that need synchronization
2. Check if Relaxed is sufficient (counters, flags with no dependent data)
3. Confirm Acquire/Release vs SeqCst choice matters (most code doesn't need SeqCst)

Common false positives:

• Relaxed on simple counters/metrics — no ordering needed for independent values
• Relaxed on boolean flags polled in a loop — the loop provides eventual visibility
• SeqCst used "for safety" — not wrong, just potentially over-synchronized

Before Submitting Review

Submission gate — Pass: Every finding uses [FILE:LINE] ISSUE_TITLE and includes the exact line (or minimal contiguous lines) that demonstrates the issue, so a reader can jump to the proof without trusting memory.

Final verification:

1. Re-read each finding and ask: "Did I verify this is actually an issue?"
2. For each finding, can you point to the specific line that proves the issue exists?
3. Would a Rust domain expert agree this is a problem, or is it a style preference?
4. Does fixing this provide real value, or is it busywork?
5. Format every finding as: [FILE:LINE] ISSUE_TITLE
6. For each finding, ask: "Does this fix existing code, or does it request entirely new code that didn't exist before?" If the latter, downgrade to Informational.
7. If this is a re-review: ONLY verify previous fixes. Do not introduce new findings.

If uncertain about any finding, either:

• Remove it from the review
• Mark it as a question rather than an issue
• Verify by reading more code context