Quality Review with AI¶
This document explains how to use AI as a reviewer (of code, docs, architecture) without giving it the final say. It is written for ConnectSoft team members using AI for quality review.
AI can be a powerful review assistant, but it should augment human review, not replace it. This guide shows how to use AI effectively for review while maintaining human oversight.
Important
Human Sign-Off Required: Critical decisions, security changes, and production pipeline modifications must have human review and final approval, not just AI approval. AI is a review assistant, not a decision-maker.
Review Levels¶
Level 1: Quick Sanity Check¶
Purpose: Quick validation of basic correctness
Use AI For: - Syntax errors - Basic style issues - Obvious bugs - Missing imports
Human Review: Minimal (spot check)
Level 2: Detailed Style/Consistency Review¶
Purpose: Ensure consistency with ConnectSoft patterns
Use AI For: - Style compliance - Pattern adherence - Terminology consistency - Cross-link validation
Human Review: Required (verify AI findings)
Level 3: Architecture & Risk Check¶
Purpose: Validate architecture and identify risks
Use AI For: - Architecture pattern compliance - Risk identification - Alternative suggestions - Best practice validation
Human Review: Required (critical decisions)
Using AI to Review Code Changes¶
Code Review Prompt Template¶
You are an expert .NET code reviewer reviewing code for ConnectSoft projects.
Goal: Review [code/file/PR] for quality and compliance.
Context:
- Code: [code snippet or file]
- Project: [project name]
- Pattern: ConnectSoft microservice template
- Libraries: ConnectSoft.Extensions.*
Review for:
- Clean Architecture compliance (Domain → Application → Infrastructure)
- DDD patterns (aggregates, entities, domain events)
- Observability (logging, tracing, metrics)
- Test coverage
- Code style and consistency
- Security issues
- Performance concerns
Output:
- Issues found (with severity)
- Suggestions for improvement
- Compliance assessment
- Risk assessment
Limitations¶
AI Cannot: - Run tests or validate functionality - Understand business context fully - Make final architectural decisions - Approve security changes - Understand customer requirements
Always: - Run tests manually - Validate functionality - Get human review for critical changes - Approve security changes with human security review
Using AI to Review Documentation¶
Documentation Review Prompt Template¶
You are the ConnectSoft Documentation Editor reviewing documentation.
Goal: Review [document] for quality and compliance.
Context:
- Document: [document content]
- Style guide: ConnectSoft documentation style guide
- Related docs: [list related docs]
Review for:
- Style compliance (title, intro, headings, callouts)
- Missing cross-links
- Inconsistent terminology
- Over-complication
- Missing sections
- Accuracy and completeness
Output:
- Style issues found
- Missing elements
- Suggested improvements
- Cross-link suggestions
- Terminology inconsistencies
Documentation Review Checklist¶
- Single
#title present - 1-3 sentence intro present
- Logical heading structure
- Appropriate callouts used
- Cross-links to related docs
- Consistent terminology
- No missing sections
- Content accurate and complete
See: Documentation Style Guide for style details.
Cross-Checking Architecture Decisions¶
Architecture Review Prompt Template¶
You are a ConnectSoft architecture reviewer.
Goal: Review architecture decision for [component/service].
Context:
- Decision: [architecture decision]
- Alternatives: [alternatives considered]
- Constraints: [technical/business constraints]
Review for:
- Alignment with Clean Architecture
- DDD pattern compliance
- Event-driven architecture compliance
- Cloud-native principles
- ConnectSoft patterns
- Risks and mitigations
- Alternatives not considered
Output:
- Architecture assessment
- Risk identification
- Alternative suggestions
- Compliance evaluation
Architecture Review Process¶
- AI Review - Use AI to review architecture decision
- Human Review - Human architect reviews AI findings
- Discussion - Discuss findings and alternatives
- Decision - Make final decision (human)
- ADR - Document decision in ADR
See: Decision Records Process for ADR process.
Human Sign-Off Rules¶
Critical Decisions Requiring Human Review¶
- Architecture Decisions - All architecture decisions require human architect review
- Security Changes - All security changes require human security review
- Production Pipeline Changes - All production pipeline changes require human DevOps review
- Breaking Changes - All breaking changes require human review and approval
- Business Decisions - All business decisions require human product/business review
Review Process¶
- AI Review - Use AI for initial review
- Human Review - Human reviewer reviews AI findings
- Discussion - Discuss findings and concerns
- Approval - Human reviewer approves or requests changes
- Documentation - Document decision if significant
Important
Human Sign-Off Required: 1. Critical Decisions - Architecture, security, production changes 2. Breaking Changes - All breaking changes 3. Business Decisions - Product and business decisions 4. Final Approval - AI provides input, humans make final decisions
Related Documents¶
- AI Collaboration Overview - AI collaboration principles
- Prompts and Patterns - Prompt templates
- Multi-Agent Workflows - Workflow patterns
- Using Cursor with ConnectSoft - Cursor usage
- Documentation Style Guide - Documentation style
- Decision Records Process - ADR process
- Security & Compliance - Security requirements