Dimension	Focus	Priority
Security	Vulnerabilities, auth, data exposure	Critical
Performance	Speed, memory, scalability bottlenecks	High
Correctness	Logic errors, edge cases, data integrity	High
Maintainability	Readability, structure, future-proofing	Medium
Testing	Coverage, quality, reliability of tests	Medium
Accessibility	WCAG compliance, keyboard nav, screen readers	Medium
Documentation	Comments, API docs, changelog entries	Low

Security Checklist

Review every change for these vulnerabilities:

Performance Checklist

Correctness Checklist

Maintainability Checklist

Testing Checklist

Test Coverage — New logic paths have corresponding tests; critical paths have both happy-path and failure-case tests
Edge Case Tests — Tests cover boundary values, empty inputs, nulls, and error conditions
No Flaky Tests — Tests are deterministic; no reliance on timing, external services, or shared mutable state
Test Independence — Each test sets up its own state and tears it down; test order does not affect results
Meaningful Assertions — Tests assert on behavior and outcomes, not implementation details
Test Readability — Tests follow Arrange-Act-Assert; test names describe the scenario and expected outcome
Mocking Discipline — Only external boundaries (network, DB, filesystem) are mocked
Regression Tests — Bug fixes include a test that reproduces the original bug and proves it is resolved

Review Process

Work through the code in three passes. Do not try to catch everything in one read.

Pass	Focus	Time	What to Look For
First	High-level structure	2-5 min	Architecture fit, file organization, API design, overall approach
Second	Line-by-line detail	Bulk	Logic errors, security issues, performance problems, edge cases
Third	Edge cases & hardening	5 min	Failure modes, concurrency, boundary values, missing tests

First Pass (2-5 minutes)

Read the PR description and linked issue
Scan the file list — does the change scope make sense?
Check the overall approach — is this the right solution to the problem?
Verify the change does not introduce architectural drift

Second Pass (bulk of review time)

Read each file diff top to bottom
Check every function change against the checklists above
Verify error handling at every I/O boundary
Flag anything that makes you pause — trust your instincts

Third Pass (5 minutes)

Think about what could go wrong in production
Check for missing tests on the code paths you flagged
Verify rollback safety — can this change be reverted without data loss?
Confirm documentation and changelog are updated if needed

Severity Levels

Classify every comment by severity so the author knows what blocks merge.

Level	Label	Meaning	Blocks Merge?
Critical	`[CRITICAL]`	Security vulnerability, data loss, or crash in production	Yes
Major	`[MAJOR]`	Bug, logic error, or significant performance regression	Yes
Minor	`[MINOR]`	Improvement that would reduce future maintenance cost	No
Nitpick	`[NIT]`	Style preference, naming suggestion, or trivial cleanup	No

Always prefix your review comment with the severity label. This removes ambiguity about what matters.

Giving Feedback

Principles

Be specific — Point to the exact line and explain the issue, not just "this is wrong"
Explain why — State the risk or consequence, not just the rule
Suggest a fix — Offer a concrete alternative or code snippet when possible
Ask, don't demand — Use questions for subjective points: "What do you think about...?"
Acknowledge good work — Call out clean solutions, clever optimizations, or thorough tests
Separate blocking from non-blocking — Use severity labels so the author knows what matters

Example Comments

Bad:

This is wrong. Fix it.

Good:

[MAJOR] This query interpolates user input directly into the SQL string (line 42), which is vulnerable to SQL injection. Consider using a parameterized query:
SELECT * FROM users WHERE id = $1

Bad:

Why didn't you add tests?

Good:

[MINOR] The new calculateDiscount() function has a few branching paths — could we add tests for the zero-quantity and negative-price edge cases to prevent regressions?

Bad:

I would have done this differently.

Good:

[NIT] This works well. An alternative approach could be extracting the retry logic into a shared withRetry() wrapper — but that's optional and could be a follow-up.

Review Anti-Patterns

Avoid these common traps that waste time and damage team trust:

Anti-Pattern	Description
Rubber-Stamping	Approving without reading. Creates false confidence and lets bugs through.
Bikeshedding	Spending 30 minutes debating a variable name while ignoring a race condition.
Blocking on Style	Refusing to approve over formatting that a linter should enforce automatically.
Gatekeeping	Requiring your personal preferred approach when the submitted one is correct.
Drive-by Reviews	Leaving one vague comment and disappearing. Commit to following through.
Scope Creep Reviews	Requesting unrelated refactors that should be separate PRs.
Stale Reviews	Letting PRs sit for days. Review within 24 hours or hand off to someone else.
Emotional Language	"This is terrible" or "obviously wrong." Critique the code, not the person.

NEVER Do

NEVER approve without reading every changed line — rubber-stamping is worse than no review
NEVER block a PR solely for style preferences — use a linter; humans review logic
NEVER leave feedback without a severity level — ambiguity causes wasted cycles
NEVER request changes without explaining why — "fix this" teaches nothing
NEVER review more than 400 lines in one sitting — comprehension drops sharply; break large PRs into sessions
NEVER skip the security checklist — one missed vulnerability outweighs a hundred style nits
NEVER make it personal — review the code, never the coder; assume good intent

code-review

Code Review Checklist

Installation

OpenClaw / Moltbot / Clawbot

Review Dimensions