AgentSkillsCN

backend-code-review

聚焦后端逻辑、功能正确性、整洁代码模式以及本地错误处理的技术审查。

SKILL.md
--- frontmatter
name: backend-code-review
description: Focused technical review of backend logic, functional correctness, clean code patterns, and local error handling.

Backend Code Review: Logic & Correctness

Scope Definition (MANDATORY)

Identify the specific code units under review (Functions, Classes, or specific Files). The reviewer must isolate the business logic from the underlying infrastructure.

Mandatory context:

  1. •Files affected.
  2. •Business objective of the code.
  3. •Local dependencies.

Audit Dimensions

1. Functional Logic & Business Rules

Verify if the code actually achieves the intended business goal without side effects.

  • •Edge Case Analysis: Are nulls, empty strings, or zero values handled?
  • •Logic Branching: Are all if/else paths valid and reachable?
  • •Output Consistency: Does it return what the caller expects in all scenarios?

2. Clean Code & Maintainability

Evaluate the "health" of the code for future developers.

  • •Naming: Are variables/functions descriptive or generic? (Avoid data, process, temp).
  • •Complexity: Is the function too long or deeply nested? (Cyclomatic complexity).
  • •Redundancy (DRY): Is there logic that should be abstracted?
  • •Declarative vs Imperative: Is the code easy to read at a glance?

3. Robust Error Handling (Local)

Focus on how the code fails.

  • •Graceful Failure: Does one small failure crash the entire process?
  • •Specific Catching: Are we catching Error (bad) or specific exceptions (good)?
  • •Status Codes: Are HTTP codes (400, 422, 500) mapped correctly to the logic error?
  • •Logging: Are enough details logged to debug, without leaking PII?

4. Input Integrity

Review the boundary where data enters the logic.

  • •Type Safety: Are types enforced (TypeScript or runtime checks)?
  • •Sanitization: Is the input cleaned before manipulation?
  • •Constraint Enforcement: Are min/max, required fields, and formats validated?

What this skill does NOT review (Avoid overlap)

  • •Architecture: Do not review layer separation or dependency direction (Use architecture-review).
  • •Persistence: Do not review SQL queries, N+1, or transaction logic (Use data-layer-review).
  • •Security/Auth: Do not review JWTs or Roles (Use auth-security-audit).

Mandatory Output Format

1. Logic Integrity Score [0-10]

Quick assessment of how likely this code is to fail in production.

2. Functional Flaws

Detailed list of logic bugs or missing edge cases.

3. Debt & Maintainability

Specific naming or complexity issues with suggested refactors.

4. Error Handling Audit

Assessment of the failure modes and logging strategy.