Deep Review: Multi-Agent Quality Improvement Framework

深度评审：多Agent质量提升框架

Execute this skill to get constructive feedback and improvement suggestions through balanced expert analysis.

执行该Skill即可通过均衡的专家分析获取建设性反馈和改进建议。

Execution Instructions

执行说明

When invoked, you will:

Resolve scope and context — invoke context skill (always), then preflight if confidence is low
Populate review scope from working_scope
Spawn reviewer agents in parallel for comprehensive feedback
Aggregate suggestions from all reviewers with proper weighting
Generate improvement report with actionable recommendations
Save report to
```
.outputs/review/
```

Note: This is a review for improvement, not pass/fail verification.

被调用时，你需要完成以下步骤：

确定范围与上下文 —— 始终调用context Skill，如果置信度较低则再执行preflight
根据working_scope填充评审范围
并行启动评审Agent以获取全面的反馈
加权聚合所有评审Agent的建议
生成包含可落地建议的改进报告
将报告保存到
```
.outputs/review/
```
目录

注意： 这是提升性质的评审，而非通过/不通过的验证。

Dependency Check

依赖检查

Before executing any step, verify all required skills are present:

[skills-root]/context/SKILL.md
[skills-root]/preflight/SKILL.md
[skills-root]/domain-registry/README.md

Where

[skills-root]

is the parent of this skill's directory. Resolve with

ls ../

from this skill's location.

If any required file is missing → stop immediately and output:

⚠ Missing required skills for deep-review:

  {missing-skill}
    Expected: {skills-root}/{missing-skill}/SKILL.md

Install the missing skill(s):
  git clone https://github.com/mikeng-io/agent-skills /tmp/agent-skills
  cp -r /tmp/agent-skills/skills/{missing-skill} {skills-root}/

Or install the full suite at once:
  cp -r /tmp/agent-skills/skills/ {skills-root}/

All dependencies present → proceed to Step 0.

执行任何步骤前，请确认所有所需Skill均已存在：

[skills-root]/context/SKILL.md
[skills-root]/preflight/SKILL.md
[skills-root]/domain-registry/README.md

其中

[skills-root]

是当前Skill目录的父目录。你可以在当前Skill目录下执行

ls ../

确认路径。

如果缺少任何所需文件 → 立即停止并输出：

⚠ Missing required skills for deep-review:

  {missing-skill}
    Expected: {skills-root}/{missing-skill}/SKILL.md

Install the missing skill(s):
  git clone https://github.com/mikeng-io/agent-skills /tmp/agent-skills
  cp -r /tmp/agent-skills/skills/{missing-skill} {skills-root}/

Or install the full suite at once:
  cp -r /tmp/agent-skills/skills/ {skills-root}/

所有依赖均已存在 → 进入步骤0。

Step 0: Scope & Context Resolution

步骤0：范围与上下文解析

Context (always required):

Invoke

Skill("context")

first. It classifies the artifact, detects domains from domain-registry, and determines routing confidence:

yaml

context_report:
  artifact_type: ""  # code | financial | marketing | creative | research | mixed
  domains: []        # matched domain names from domain-registry
  routing: ""        # parallel-workflow | debate-protocol | deep-council
  confidence: ""     # high | medium | low

Preflight (conditional — triggered by context confidence):

Invoke

Skill("preflight")

only if

context_report.confidence == "low"

OR one or more signals remain unresolved:

Artifact is not clearly identified
Intent is ambiguous (what aspect to improve?)
Domains could not be detected
Scope is too broad

Preflight fills exactly the gaps context could not resolve (max 3 questions, one at a time):

yaml

scope_clarification:
  artifact: ""       # what to review
  intent: "review"
  domains: []        # supplements context_report.domains
  constraints: []    # explicit areas to focus on (e.g., "performance", "security")
  confidence: ""     # high | medium

If

context_report.confidence == "high"

→ skip preflight entirely.

Merge into working scope:

yaml

working_scope:
  artifact: ""            # files, topics, or description of what to review
  domains: []             # from context_report (authoritative), supplemented by preflight
  concerns: []            # from context signals and scope_clarification.constraints
  context_summary: ""     # combined description for reviewer agent prompts

Use

working_scope

throughout this skill.

上下文（始终需要）：

首先调用

Skill("context")

。它会对交付物进行分类，从domain-registry中检测所属领域，并确定路由置信度：

yaml

context_report:
  artifact_type: ""  # code | financial | marketing | creative | research | mixed
  domains: []        # matched domain names from domain-registry
  routing: ""        # parallel-workflow | debate-protocol | deep-council
  confidence: ""     # high | medium | low

Preflight（条件执行 —— 由上下文置信度触发）：

仅当

context_report.confidence == "low"

或存在一个及以上未解决的信号时，才调用

Skill("preflight")

：

交付物未明确识别
意图不明确（需要改进哪些方面？）
无法检测所属领域
范围过宽

Preflight仅会填补上下文无法解决的空缺（最多3个问题，一次询问一个）：

yaml

scope_clarification:
  artifact: ""       # what to review
  intent: "review"
  domains: []        # supplements context_report.domains
  constraints: []    # explicit areas to focus on (e.g., "performance", "security")
  confidence: ""     # high | medium

如果

context_report.confidence == "high"

→ 完全跳过preflight。

合并到工作范围：

yaml

working_scope:
  artifact: ""            # files, topics, or description of what to review
  domains: []             # from context_report (authoritative), supplemented by preflight
  concerns: []            # from context signals and scope_clarification.constraints
  context_summary: ""     # combined description for reviewer agent prompts

整个Skill执行过程中均使用该

working_scope

。

Step 1: Populate Review Scope

步骤1：填充评审范围

Using

working_scope

from Step 0, populate the review context:

yaml

review_context:
  files: []              # from working_scope.artifact
  artifacts: []          # additional artifacts from working_scope
  topics: []             # key topics from context_report
  concerns: []           # from working_scope.concerns
  intent: ""             # from working_scope — what user wants to improve
  domain_inference: []   # from working_scope.domains

使用步骤0得到的

working_scope

填充评审上下文：

yaml

review_context:
  files: []              # from working_scope.artifact
  artifacts: []          # additional artifacts from working_scope
  topics: []             # key topics from context_report
  concerns: []           # from working_scope.concerns
  intent: ""             # from working_scope — what user wants to improve
  domain_inference: []   # from working_scope.domains

Step 2: Spawn Reviewer Agents in Parallel

步骤2：并行启动评审Agent

Spawn reviewer sub-agents in parallel using the Task tool.

使用Task工具并行启动评审子Agent。

Reviewer Distribution

评审员分配规则

yaml

reviewer_selection:
  always_spawn:
    - Best Practices Expert (weight varies by domain)
    - Alternative Approaches Expert

  domain_driven_spawn:
    - Read domain-registry to select domain experts matching conversation signals
    - Each selected domain adds one domain expert reviewer
    - Replace Code Quality Reviewer with domain-appropriate quality reviewer
      (e.g., financial → Financial Accuracy Reviewer, design → Visual Quality Reviewer)

  fallback_if_no_domain_match:
    - Code Quality Reviewer (30% weight)
    - Performance Optimizer (15% weight)

execution:
  mode: parallel
  max_concurrent: 4
  capability: high

yaml

reviewer_selection:
  always_spawn:
    - Best Practices Expert (weight varies by domain)
    - Alternative Approaches Expert

  domain_driven_spawn:
    - Read domain-registry to select domain experts matching conversation signals
    - Each selected domain adds one domain expert reviewer
    - Replace Code Quality Reviewer with domain-appropriate quality reviewer
      (e.g., financial → Financial Accuracy Reviewer, design → Visual Quality Reviewer)

  fallback_if_no_domain_match:
    - Code Quality Reviewer (30% weight)
    - Performance Optimizer (15% weight)

execution:
  mode: parallel
  max_concurrent: 4
  capability: high

Agent Templates

Agent模板

Best Practices Expert

最佳实践专家

Weight: 35%
Purpose: Suggest industry best practices and standards
Capability: high

You are a BEST PRACTICES EXPERT. Your role is to suggest improvements based on industry standards and best practices.

Weight: 35%
Purpose: Suggest industry best practices and standards
Capability: high

You are a BEST PRACTICES EXPERT. Your role is to suggest improvements based on industry standards and best practices.

Your Mindset

"This works, but here's how to make it follow best practices and be more maintainable."

Focus Areas

Industry standards and conventions
Framework/language-specific best practices
Design principles (SOLID, DRY, KISS, etc.)
Security best practices
Accessibility standards (if applicable)
Testing best practices

Industry standards and conventions
Framework/language-specific best practices
Design principles (SOLID, DRY, KISS, etc.)
Security best practices
Accessibility standards (if applicable)
Testing best practices

Context to Review

{conversation_context}

Your Scope

{scope_description}

Output Format (JSON)

{ "agent": "best-practices", "suggestions": [ { "category": "Security | Architecture | Testing | Documentation | etc.", "severity": "CRITICAL | HIGH | MEDIUM | LOW", "current_approach": "What's being done now", "best_practice": "What the industry standard is", "suggestion": "Specific improvement to make", "rationale": "Why this is better", "example": "Code example or reference (if applicable)", "resources": ["Links to documentation, standards, guides"] } ], "overall_assessment": "General feedback on alignment with best practices" }

undefined

{ "agent": "best-practices", "suggestions": [ { "category": "Security | Architecture | Testing | Documentation | etc.", "severity": "CRITICAL | HIGH | MEDIUM | LOW", "current_approach": "What's being done now", "best_practice": "What the industry standard is", "suggestion": "Specific improvement to make", "rationale": "Why this is better", "example": "Code example or reference (if applicable)", "resources": ["Links to documentation, standards, guides"] } ], "overall_assessment": "General feedback on alignment with best practices" }

undefined

Code Quality Reviewer

代码质量评审员

Weight: 30%
Purpose: Improve code quality, readability, and maintainability
Capability: high

You are a CODE QUALITY REVIEWER. Your role is to suggest improvements for readability, maintainability, and code health.

Weight: 30%
Purpose: Improve code quality, readability, and maintainability
Capability: high

You are a CODE QUALITY REVIEWER. Your role is to suggest improvements for readability, maintainability, and code health.

Your Mindset

"This code works, but here's how to make it clearer, more maintainable, and easier to work with."

Focus Areas

Code readability and clarity
Naming conventions
Function/method size and complexity
Code organization and structure
Documentation and comments
Error handling patterns
Code duplication (DRY violations)
Magic numbers/strings

Code readability and clarity
Naming conventions
Function/method size and complexity
Code organization and structure
Documentation and comments
Error handling patterns
Code duplication (DRY violations)
Magic numbers/strings

Context to Review

{conversation_context}

Output Format (JSON)

{ "agent": "code-quality", "suggestions": [ { "category": "Readability | Maintainability | Organization | Documentation", "severity": "CRITICAL | HIGH | MEDIUM | LOW", "location": "File path and line number (if applicable)", "issue": "What could be improved", "suggestion": "Specific improvement", "before": "Current code pattern (if applicable)", "after": "Improved code pattern (if applicable)", "impact": "How this improves code quality" } ], "code_health_score": "Assessment of overall code health", "positive_aspects": ["What's already good"] }

undefined

{ "agent": "code-quality", "suggestions": [ { "category": "Readability | Maintainability | Organization | Documentation", "severity": "CRITICAL | HIGH | MEDIUM | LOW", "location": "File path and line number (if applicable)", "issue": "What could be improved", "suggestion": "Specific improvement", "before": "Current code pattern (if applicable)", "after": "Improved code pattern (if applicable)", "impact": "How this improves code quality" } ], "code_health_score": "Assessment of overall code health", "positive_aspects": ["What's already good"] }

undefined

Alternative Approaches Expert

替代方案专家

Weight: 20%
Purpose: Suggest different approaches and trade-offs
Capability: high

You are an ALTERNATIVE APPROACHES EXPERT. Your role is to present different ways to solve the same problem with trade-off analysis.

Weight: 20%
Purpose: Suggest different approaches and trade-offs
Capability: high

You are an ALTERNATIVE APPROACHES EXPERT. Your role is to present different ways to solve the same problem with trade-off analysis.

Your Mindset

"The current approach works, but here are alternative solutions with their pros and cons."

Focus Areas

Different design patterns
Alternative architectures
Different technology choices
Simpler solutions
More scalable approaches
Different frameworks/libraries
Trade-offs between approaches

Different design patterns
Alternative architectures
Different technology choices
Simpler solutions
More scalable approaches
Different frameworks/libraries
Trade-offs between approaches

Context to Review

{conversation_context}

Output Format (JSON)

{ "agent": "alternative-approaches", "alternatives": [ { "name": "Name of alternative approach", "description": "What this approach involves", "pros": ["Advantages of this approach"], "cons": ["Disadvantages of this approach"], "when_to_use": "Scenarios where this is better", "complexity": "HIGH | MEDIUM | LOW", "example": "Code example or reference (if applicable)" } ], "current_approach_assessment": { "strengths": ["What's good about current approach"], "weaknesses": ["What could be better"], "verdict": "When current approach is appropriate" } }

undefined

{ "agent": "alternative-approaches", "alternatives": [ { "name": "Name of alternative approach", "description": "What this approach involves", "pros": ["Advantages of this approach"], "cons": ["Disadvantages of this approach"], "when_to_use": "Scenarios where this is better", "complexity": "HIGH | MEDIUM | LOW", "example": "Code example or reference (if applicable)" } ], "current_approach_assessment": { "strengths": ["What's good about current approach"], "weaknesses": ["What could be better"], "verdict": "When current approach is appropriate" } }

undefined

Performance Optimizer

性能优化专家

Weight: 15%
Purpose: Identify performance optimization opportunities
Capability: high

You are a PERFORMANCE OPTIMIZER. Your role is to identify opportunities for performance improvements.

Weight: 15%
Purpose: Identify performance optimization opportunities
Capability: high

You are a PERFORMANCE OPTIMIZER. Your role is to identify opportunities for performance improvements.

Your Mindset

"This works, but here's how to make it faster, more efficient, or more scalable."

Focus Areas

Algorithm complexity (Big O)
Database query optimization
Caching opportunities
Lazy loading vs eager loading
Resource utilization (memory, CPU, network)
Bottlenecks and hot paths
Scalability considerations
Frontend performance (if applicable)

Algorithm complexity (Big O)
Database query optimization
Caching opportunities
Lazy loading vs eager loading
Resource utilization (memory, CPU, network)
Bottlenecks and hot paths
Scalability considerations
Frontend performance (if applicable)

Context to Review

{conversation_context}

Output Format (JSON)

{ "agent": "performance", "optimizations": [ { "category": "Algorithm | Database | Caching | Resource | Scalability", "severity": "CRITICAL | HIGH | MEDIUM | LOW", "current_complexity": "O(n^2), 500ms response time, etc.", "opportunity": "What can be optimized", "suggestion": "Specific optimization", "expected_improvement": "How much faster/better", "trade_offs": ["What you give up for this optimization"], "effort": "HIGH | MEDIUM | LOW" } ], "performance_assessment": "Overall performance analysis", "premature_optimization_warning": "Areas where optimization might not be worth it" }

---

{ "agent": "performance", "optimizations": [ { "category": "Algorithm | Database | Caching | Resource | Scalability", "severity": "CRITICAL | HIGH | MEDIUM | LOW", "current_complexity": "O(n^2), 500ms response time, etc.", "opportunity": "What can be optimized", "suggestion": "Specific optimization", "expected_improvement": "How much faster/better", "trade_offs": ["What you give up for this optimization"], "effort": "HIGH | MEDIUM | LOW" } ], "performance_assessment": "Overall performance analysis", "premature_optimization_warning": "Areas where optimization might not be worth it" }

---

Step 3: Aggregate Suggestions

步骤3：聚合建议

After all reviewer agents complete, aggregate their suggestions:

所有评审Agent执行完成后，聚合它们的建议：

Categorize by Priority

按优先级分类

yaml

high_priority:
  - Suggestions marked as HIGH priority
  - Security concerns from best practices
  - Critical code quality issues

medium_priority:
  - Suggestions marked as MEDIUM priority
  - Maintainability improvements
  - Alternative approaches to consider

low_priority:
  - Nice-to-have improvements
  - Minor optimizations
  - Style preferences

yaml

high_priority:
  - Suggestions marked as HIGH priority
  - Security concerns from best practices
  - Critical code quality issues

medium_priority:
  - Suggestions marked as MEDIUM priority
  - Maintainability improvements
  - Alternative approaches to consider

low_priority:
  - Nice-to-have improvements
  - Minor optimizations
  - Style preferences

Identify Common Themes

识别共同主题

Look for suggestions mentioned by multiple reviewers:

If 2+ reviewers mention same issue → Highlight as important
If reviewers conflict → Present both viewpoints
If reviewers agree → Emphasize consensus

查找多个评审员都提到的建议：

如果2名及以上评审员提到同一个问题 → 标记为重要
如果评审员意见冲突 → 同时呈现两种观点
如果评审员意见一致 → 强调共识

Build Summary Table

构建汇总表

Aspect	Assessment	Key Suggestions
Best Practices	Strong/Moderate/Weak	Top 3 suggestions
Code Quality	Score/10	Top 3 improvements
Architecture	Appropriate/Consider Alternatives	Alternative approaches
Performance	Good/Needs Optimization	Top optimizations

维度	评估结果	核心建议
最佳实践	优秀/中等/不足	前3条建议
代码质量	得分/10	前3项改进点
架构	合理/可考虑替代方案	替代方案
性能	良好/需要优化	核心优化点

Step 4: Generate Review Report

步骤4：生成评审报告

Generate a markdown report with this structure:

markdown

undefined

按照以下结构生成markdown报告：

markdown

undefined

Deep Review Report

深度评审报告

Review Type: Quality Improvement Reviewed At: {timestamp} Scope: {what_was_reviewed} Reviewers: 4 expert agents

评审类型： 质量提升 评审时间： {timestamp} 评审范围： {what_was_reviewed} 评审员： 4名专家Agent

Executive Summary

执行摘要

{2-3 paragraphs summarizing key findings and recommendations}

Overall Assessment: {High quality / Good with room for improvement / Needs work}

Top 3 Recommendations:

{Most important suggestion}
{Second most important}
{Third most important}

{2-3段总结核心发现和建议}

整体评估： {高质量 / 良好，仍有改进空间 / 需要优化}

前3项建议：

{最重要的建议}
{第二重要的建议}
{第三重要的建议}

Review Summary

评审汇总

Aspect	Assessment	Priority Suggestions
Best Practices	{assessment}	{count} suggestions
Code Quality	{score}/10	{count} improvements
Alternatives	{count} options	{count} trade-offs
Performance	{assessment}	{count} optimizations

维度	评估结果	优先级建议
最佳实践	{assessment}	{count} 条建议
代码质量	{score}/10	{count} 项改进
替代方案	{count} 个选项	{count} 项权衡
性能	{assessment}	{count} 项优化

High Priority Suggestions

高优先级建议

{Category}: {Suggestion Title}

{类别}: {建议标题}

Priority: HIGH Suggested by: {Agent name(s)}

Current Approach: {What's being done now}

Suggestion: {Specific improvement to make}

Rationale: {Why this is important}

Example:

// Before
{current_code_pattern}

// After
{improved_code_pattern}

Impact: {Expected benefit}

{Repeat for each high-priority suggestion}

优先级： 高 提出方： {Agent名称}

当前方案： {当前实现内容}

建议： {具体改进措施}

理由： {该改进的重要性说明}

示例：

// 优化前
{current_code_pattern}

// 优化后
{improved_code_pattern}

影响： {预期收益}

{重复展示所有高优先级建议}

Medium Priority Suggestions

中优先级建议

{Same format as high priority, grouped by category}

{格式与高优先级一致，按类别分组}

Alternative Approaches

替代方案

Alternative 1: {Approach Name}

替代方案1：{方案名称}

Description: {What this involves}

Pros:

{Advantage 1}
{Advantage 2}

Cons:

{Disadvantage 1}
{Disadvantage 2}

When to Use: {Scenarios where this is better}

Complexity: {HIGH/MEDIUM/LOW}

{Repeat for each alternative}

描述： {方案内容说明}

优点：

{优势1}
{优势2}

缺点：

{劣势1}
{劣势2}

适用场景： {该方案更适合的场景}

复杂度： {高/中/低}

{重复展示所有替代方案}

Performance Optimization Opportunities

性能优化机会

{Optimization Title}

{优化项标题}

Category: {Algorithm/Database/Caching/etc.} Priority: {HIGH/MEDIUM/LOW} Effort: {HIGH/MEDIUM/LOW}

Current Performance: {Metrics or complexity}

Optimization: {Specific suggestion}

Expected Improvement: {How much better}

Trade-offs:

{What you give up}

{Repeat for each optimization}

类别： {算法/数据库/缓存/其他} 优先级： {高/中/低} 改造成本： {高/中/低}

当前性能： {指标或复杂度}

优化方案： {具体建议}

预期提升： {优化后的收益}

权衡：

{需要牺牲的方面}

{重复展示所有优化项}

Positive Aspects

优点

What's Already Good:

{Positive aspect 1}
{Positive aspect 2}
{Positive aspect 3}

Strengths to Maintain:

{Strength 1}
{Strength 2}

已做的较好的方面：

{优点1}
{优点2}
{优点3}

需要保持的优势：

{优势1}
{优势2}

Resources & References

资源与参考

Best Practices:

{Link to standard/guide}
{Link to documentation}

Alternative Approaches:

{Link to pattern description}
{Link to comparison}

Performance:

{Link to optimization guide}
{Link to benchmarking tool}

最佳实践：

{标准/指南链接}
{文档链接}

替代方案：

{模式说明链接}
{对比分析链接}

性能：

{优化指南链接}
{基准测试工具链接}

Next Steps

后续步骤

Recommended Action Plan:

Immediate (High Priority):
- {Action item 1}
- {Action item 2}
Short Term (Medium Priority):
- {Action item 3}
- {Action item 4}
Long Term (Low Priority):
- {Action item 5}
- {Action item 6}

Estimated Impact:

Code Quality: {improvement estimate}
Maintainability: {improvement estimate}
Performance: {improvement estimate}

---

推荐行动计划：

立即执行（高优先级）：
- {行动项1}
- {行动项2}
短期执行（中优先级）：
- {行动项3}
- {行动项4}
长期执行（低优先级）：
- {行动项5}
- {行动项6}

预期影响：

代码质量：{提升幅度预估}
可维护性：{提升幅度预估}
性能：{提升幅度预估}

---

Step 5: Save Report

步骤5：保存报告

Artifact Output

交付物输出

Save to

.outputs/review/{YYYYMMDD-HHMMSS}-review-{slug}.md

with YAML frontmatter:

yaml

---
skill: deep-review
timestamp: {ISO-8601}
artifact_type: review
domains: [{domain1}, {domain2}]
quality_assessment: "High quality | Good with room for improvement | Needs work"
context_summary: "{brief description of what was reviewed}"
session_id: "{unique id}"
---

Also save JSON companion:

{timestamp}-review-{slug}.json

No symlinks. To find the latest artifact:

bash

ls -t .outputs/review/ | head -1

QMD Integration (optional, progressive enhancement):

bash

qmd collection add .outputs/review/ --name "deep-review-artifacts" --mask "**/*.md" 2>/dev/null || true
qmd update 2>/dev/null || true

Output Structure:

.outputs/review/
├── 20260130-143000-review-report.md
└── 20260130-143000-review-report.json

保存到

.outputs/review/{YYYYMMDD-HHMMSS}-review-{slug}.md

，包含以下YAML前置元数据：

yaml

---
skill: deep-review
timestamp: {ISO-8601}
artifact_type: review
domains: [{domain1}, {domain2}]
quality_assessment: "High quality | Good with room for improvement | Needs work"
context_summary: "{brief description of what was reviewed}"
session_id: "{unique id}"
---

同时保存JSON配套文件：

{timestamp}-review-{slug}.json

不要使用软链接。 查找最新交付物的命令：

bash

ls -t .outputs/review/ | head -1

QMD集成（可选，渐进式增强）：

bash

qmd collection add .outputs/review/ --name "deep-review-artifacts" --mask "**/*.md" 2>/dev/null || true
qmd update 2>/dev/null || true

输出结构：

.outputs/review/
├── 20260130-143000-review-report.md
└── 20260130-143000-review-report.json

Configuration (Optional)

配置（可选）

yaml

undefined

yaml

undefined

.outputs/review/config.yaml

review:

Reviewer weights

weights: best_practices: 0.35 code_quality: 0.30 alternatives: 0.20 performance: 0.15

Priority thresholds

high_priority_threshold: 0.8 medium_priority_threshold: 0.5

Output options

include_code_examples: true include_resources: true max_suggestions_per_category: 10


**Environment Variables:**
```bash
export DEEP_REVIEW_OUTPUT_DIR=".outputs/review/"
export DEEP_REVIEW_INCLUDE_EXAMPLES="true"

review:

评审员权重

weights: best_practices: 0.35 code_quality: 0.30 alternatives: 0.20 performance: 0.15

优先级阈值

high_priority_threshold: 0.8 medium_priority_threshold: 0.5

输出选项

include_code_examples: true include_resources: true max_suggestions_per_category: 10


**环境变量：**
```bash
export DEEP_REVIEW_OUTPUT_DIR=".outputs/review/"
export DEEP_REVIEW_INCLUDE_EXAMPLES="true"

Notes

注意事项

Constructive Focus: This is about improvement, not criticism
No Verdict: No pass/fail - only suggestions
Actionable: All suggestions include specific actions
Balanced: Includes positive aspects, not just problems
Conversation-Driven: Extracts context from what was discussed
Domain-Agnostic: Works for any domain (code, design, content, etc.)
Parallel Execution: All reviewers run simultaneously for speed
Multi-Model: For cross-model review confidence, see
```
deep-council
```
Domain-Aware: Reviewer distribution adapts to detected domains via domain-registry
Context Routing: If the artifact is complex or multi-domain, invoke the
```
context
```
skill first to classify artifact type and determine optimal routing (parallel-workflow vs debate-protocol vs deep-council)
DeepWiki (optional): For code artifacts, invoke
```
Skill("deepwiki")
```
before spawning reviewers if the codebase has a Devin-indexed wiki — provides architectural context that improves domain expert quality. Non-blocking; skip if unavailable.

聚焦建设性： 评审目的是改进，而非批评
无最终判定： 没有通过/不通过的结论，仅提供建议
可落地： 所有建议都包含具体执行动作
均衡性： 包含优点反馈，而非仅指出问题
对话驱动： 从对话内容中提取上下文
领域无关： 适用于任意领域（代码、设计、内容等）
并行执行： 所有评审员同时运行，提升速度
多模型： 如需跨模型评审置信度，参考
```
deep-council
```
领域感知： 评审员分配会通过domain-registry适配检测到的领域
上下文路由： 如果交付物复杂或跨领域，先调用
```
context
```
Skill对交付物类型分类，确定最优路由（并行工作流/辩论协议/深度委员会）
DeepWiki（可选）： 对于代码交付物，如果代码库有Devin索引的wiki，可以在启动评审员前调用
```
Skill("deepwiki")
```
—— 提供架构上下文可以提升领域专家评审质量。非阻塞，不可用则跳过。

deep-review

Original

Translation

Deep Review: Multi-Agent Quality Improvement Framework

深度评审：多Agent质量提升框架

Execution Instructions

执行说明

Dependency Check

依赖检查

Step 0: Scope & Context Resolution

步骤0：范围与上下文解析

Step 1: Populate Review Scope

步骤1：填充评审范围

Step 2: Spawn Reviewer Agents in Parallel

步骤2：并行启动评审Agent

Reviewer Distribution

评审员分配规则

Agent Templates

Agent模板

Best Practices Expert

最佳实践专家

Your Mindset

Your Mindset

Focus Areas

Focus Areas

Context to Review

Context to Review

Your Scope

Your Scope

Output Format (JSON)

Output Format (JSON)

Code Quality Reviewer

代码质量评审员

Your Mindset

Your Mindset

Focus Areas

Focus Areas

Context to Review

Context to Review

Output Format (JSON)

Output Format (JSON)

Alternative Approaches Expert

替代方案专家

Your Mindset

Your Mindset

Focus Areas

Focus Areas

Context to Review

Context to Review

Output Format (JSON)

Output Format (JSON)

Performance Optimizer

性能优化专家

Your Mindset

Your Mindset

Focus Areas

Focus Areas

Context to Review

Context to Review

Output Format (JSON)

Output Format (JSON)

Step 3: Aggregate Suggestions

步骤3：聚合建议

Categorize by Priority

按优先级分类

Identify Common Themes

识别共同主题

Build Summary Table

构建汇总表

Step 4: Generate Review Report

步骤4：生成评审报告

Deep Review Report

深度评审报告

Executive Summary

执行摘要

Review Summary

评审汇总

High Priority Suggestions

高优先级建议

{Category}: {Suggestion Title}