CodeReview

CodeReview

Structured methodology for evaluating code changes. Provides the framework that /simplify agents and human reviewers use to assess quality.

Customization

Before executing, check for user customizations at: ${PAI_USER_DIR}/SKILLCUSTOMIZATIONS/CodeReview/

Workflow Routing

The 5-Axis Review Framework

Every code review evaluates across five dimensions:

1. Correctness

• Does the code match the requirements/spec?
• Are edge cases handled (empty, null, boundary, concurrent)?
• Do tests cover the new behavior?
• Are error paths handled, not just happy path?

2. Readability

• Can another engineer understand this without explanation?
• Are names descriptive (not data, temp, result, x)?
• Is the logic flow obvious or does it require mental gymnastics?
• Are complex sections commented with WHY, not WHAT?

3. Architecture

• Does it fit the existing patterns in this codebase?
• Are module boundaries respected?
• Are abstractions appropriate (not premature, not missing)?
• Would this scale to 10x usage without redesign?

4. Security

• Are external inputs validated at the boundary?
• Are secrets kept out of code and logs?
• Is authentication/authorization checked where needed?
• Could this introduce XSS, injection, or CSRF?

5. Performance

• Any N+1 queries or unbounded fetches?
• Any unnecessary synchronous operations that could be async?
• Any memory leaks (uncleaned listeners, intervals, refs)?
• Would this degrade under load?

Severity Classification

Every finding gets exactly one label:

Rule: No more than 1-2 Critical per review. If you're finding 5+ Criticals, the change needs a design discussion, not a code review.

Named Simplification Patterns

When suggesting improvements, use named patterns so the author knows exactly what you mean:

Approval Philosophy

"Approve when it definitely improves overall code health, even if it isn't perfect."

The goal is continuous improvement, not perfection. Don't block a merge for Nits. Don't request a rewrite when the change is directionally correct. "I'll fix it later" is a valid response to a Nit — "I'll fix it later" is NOT valid for Critical or Important.

Examples

Example 1: Review a PR

User: "Review the changes in the last commit"
→ Invokes Review workflow
→ git diff → read tests first → walk 5 axes → classify findings
→ Output: summary with Critical/Important/Nit labels

Example 2: Pre-merge quality check

User: "Is this ready to merge?"
→ Invokes Review workflow
→ 0 Critical + 0 Important = Approve
→ 2 Nits noted as optional improvements

Example 3: Architecture concern

User: "Review this refactor for design issues"
→ Invokes Review workflow with focus on axis 3 (Architecture)
→ Checks pattern consistency, module boundaries, abstraction level

ShellCheck — Shell Script Analysis

Static analysis for shell scripts. Catches common bugs, portability issues, and style problems.

# Lint shell scripts
shellcheck script.sh

# JSON output for programmatic analysis
shellcheck -f json *.sh

# Check all shell scripts in project
find . -name "*.sh" -exec shellcheck {} +

# Specific severity
shellcheck --severity=warning script.sh

When to use:

• Code review includes shell scripts — run ShellCheck as part of the review
• CI/CD pipeline scripts — these often have subtle bugs (unquoted variables, missing error handling)
• Dockerfiles with RUN blocks — extract the shell commands and lint them
• Any .sh, .bash, or shebang-detected scripts in the changeset

Common findings:

• SC2086 — unquoted variable expansion (word splitting / globbing risk)
• SC2046 — unquoted command substitution
• SC2034 — unused variable
• SC2155 — declare and assign separately to avoid masking return values
• SC2164 — cd without || exit (failure to change directory silently continues)

Integration with 5-axis review: ShellCheck findings map to axis 1 (Correctness) and axis 4 (Security). Unquoted variables are both correctness bugs and potential injection vectors.

Integration

Works with:

• /simplify — invokes 3 review agents. This skill provides the framework they should follow.
• Algorithm VERIFY — code-producing runs invoke /simplify. This skill can be selected as a capability for deeper reviews.
• Refactor — automated tool-based scanning. Different from CodeReview (judgment-based evaluation).
• Security:SecureCoding — handles axis 4 (security) in depth. CodeReview does a quick security check; SecureCoding does a thorough one.
• TDD — handles test quality. CodeReview checks "are there tests?"; TDD ensures they're well-structured.