consensus-voting

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Consensus Voting Skill

共识投票Skill

Purpose

用途

Execute weighted multi-agent voting for critical decisions where domain expertise matters. Unlike debate (collaborative synthesis) or n-version (parallel generation), this skill focuses on structured voting with expertise weighting for security, authentication, and data-handling changes.

针对需要考量领域专业度的关键决策，执行加权多Agent投票。与辩论（协作合成）或N版本（并行生成）不同，该Skill专注于为安全、身份认证和数据处理变更提供基于专业度加权的结构化投票机制。

Key Insight from Pattern Analysis

模式分析关键洞察

From

~/.amplihack/.claude/context/DISCOVERIES.md

(Pattern Applicability Analysis):

Voting vs Expert Judgment Selection Criteria

When Voting Works:

Adversarial environment (can't trust individual nodes)

Binary or simple discrete choices

No objective quality metric available

Consensus more valuable than correctness

When Expert Judgment Works:

Cooperative environment (honest actors)

Complex quality dimensions

Objective evaluation criteria exist

Correctness more valuable than consensus

This skill uses BOTH strategically: Expert judgment to evaluate, weighted voting to decide.

来自

~/.amplihack/.claude/context/DISCOVERIES.md

（模式适用性分析）：

投票 vs 专家判断选择标准

投票适用场景：

对抗性环境（无法信任单个节点）

二元或简单离散选择

无客观质量衡量标准

共识比正确性更重要

专家判断适用场景：

协作性环境（参与者诚实可信）

复杂质量维度

存在客观评估标准

正确性比共识更重要

本Skill战略性地结合了两者：用专家判断进行评估，用加权投票做出决策。

When to Use This Skill

何时使用本Skill

AUTO-TRIGGERS (high-risk domains):

Security implementations (authentication, authorization)
Encryption and cryptographic code
Sensitive data handling (PII, credentials)
Permission and access control changes
Critical algorithm implementations

EXPLICIT TRIGGERS:

Major design decisions with competing approaches
When stakeholder buy-in matters
Binary or discrete choices needing validation
Risk-mitigating decisions for production code

AVOID FOR:

Complex trade-off analysis (use
```
debate-workflow
```
instead)
Code generation (use
```
n-version-workflow
```
instead)
Simple implementation choices
Subjective quality assessments

自动触发场景（高风险领域）：

安全实现（身份认证、授权）
加密与密码学代码
敏感数据处理（PII、凭证）
权限与访问控制变更
关键算法实现

手动触发场景：

存在竞争方案的重大设计决策
需要利益相关方认可的决策
需要验证的二元或离散选择
生产代码的风险缓解决策

避免使用场景：

复杂权衡分析（改用
```
debate-workflow
```
）
代码生成（改用
```
n-version-workflow
```
）
简单实现选择
主观质量评估

Configuration

配置

Voting Configuration

投票配置

Voting Mode:

```
simple-majority
```
- 50%+ votes to pass
```
supermajority
```
- 66%+ votes to pass (DEFAULT for security)
```
unanimous
```
- 100% agreement required

Agent Selection:

```
auto
```
- Select agents based on detected domain (DEFAULT)
```
manual
```
- Specify agents explicitly
```
comprehensive
```
- All relevant agents vote

Weight Calibration:

```
static
```
- Fixed weights per domain (DEFAULT)

投票模式：

```
simple-majority
```
- 超过50%投票即可通过
```
supermajority
```
- 超过66%投票方可通过（安全场景默认值）
```
unanimous
```
- 需要100%同意

Agent选择：

```
auto
```
- 根据检测到的领域自动选择Agent（默认值）
```
manual
```
- 明确指定Agent
```
comprehensive
```
- 所有相关Agent参与投票

权重校准：

```
static
```
- 各领域使用固定权重（默认值）

Domain Expertise Weights

领域专业度权重

Agent	Security	Auth	Data	Algorithm	General
security	3.0	2.5	2.5	1.5	1.0
reviewer	1.5	1.5	1.5	2.0	2.0
architect	1.5	2.0	2.0	2.5	2.0
tester	1.0	1.5	1.5	2.0	1.5
optimizer	0.5	0.5	0.5	2.5	1.0
cleanup	0.5	0.5	0.5	1.0	1.5

Weight Interpretation:

3.0 = Domain expert (vote counts triple)
2.0 = Significant expertise (vote counts double)
1.0 = General competence (standard vote)
0.5 = Peripheral relevance (half vote)

Agent	安全	身份认证	数据	算法	通用
security	3.0	2.5	2.5	1.5	1.0
reviewer	1.5	1.5	1.5	2.0	2.0
architect	1.5	2.0	2.0	2.5	2.0
tester	1.0	1.5	1.5	2.0	1.5
optimizer	0.5	0.5	0.5	2.5	1.0
cleanup	0.5	0.5	0.5	1.0	1.5

权重说明：

3.0 = 领域专家（投票权重为3倍）
2.0 = 具备重要专业度（投票权重为2倍）
1.0 = 通用能力（标准投票权重）
0.5 = 相关性较低（投票权重为0.5倍）

Execution Process

执行流程

Step 1: Detect Decision Domain

步骤1：检测决策领域

Analyze the change or decision to determine primary domain:

Use analyzer agent to examine code/proposal
Detect keywords: auth, encrypt, password, token, permission, credential
Identify file paths: auth/, security/, crypto/
Classify: SECURITY | AUTH | DATA | ALGORITHM | GENERAL

Domain Detection Triggers:

SECURITY: encrypt, decrypt, hash, salt, vulnerability, CVE, injection
AUTH: login, logout, session, token, jwt, oauth, password, credential
DATA: pii, gdpr, sensitive, personal, private, secret, key
ALGORITHM: sort, search, calculate, compute, process, transform

分析变更或决策内容，确定主要领域：

使用分析Agent检查代码/提案
检测关键词：auth、encrypt、password、token、permission、credential
识别文件路径：auth/、security/、crypto/
分类：SECURITY | AUTH | DATA | ALGORITHM | GENERAL

领域检测触发词：

SECURITY: encrypt, decrypt, hash, salt, vulnerability, CVE, injection
AUTH: login, logout, session, token, jwt, oauth, password, credential
DATA: pii, gdpr, sensitive, personal, private, secret, key
ALGORITHM: sort, search, calculate, compute, process, transform

Step 2: Select Voting Agents

步骤2：选择投票Agent

Based on detected domain, select relevant agents with weights:

For SECURITY domain:

security agent (weight: 3.0) - Primary expert
architect agent (weight: 1.5) - System design perspective
reviewer agent (weight: 1.5) - Code quality check
tester agent (weight: 1.0) - Testability assessment

For AUTH domain:

security agent (weight: 2.5) - Security implications
architect agent (weight: 2.0) - Integration patterns
reviewer agent (weight: 1.5) - Implementation quality
tester agent (weight: 1.5) - Auth flow testing

For DATA domain:

security agent (weight: 2.5) - Data protection
architect agent (weight: 2.0) - Data architecture
reviewer agent (weight: 1.5) - Handling patterns
tester agent (weight: 1.5) - Data validation

For ALGORITHM domain:

optimizer agent (weight: 2.5) - Performance analysis
architect agent (weight: 2.5) - Design patterns
tester agent (weight: 2.0) - Correctness testing
reviewer agent (weight: 2.0) - Code quality

根据检测到的领域，选择带权重的相关Agent：

SECURITY领域：

security agent（权重：3.0） - 主要专家
architect agent（权重：1.5） - 系统设计视角
reviewer agent（权重：1.5） - 代码质量检查
tester agent（权重：1.0） - 可测试性评估

AUTH领域：

security agent（权重：2.5） - 安全影响评估
architect agent（权重：2.0） - 集成模式
reviewer agent（权重：1.5） - 实现质量
tester agent（权重：1.5） - 认证流程测试

DATA领域：

security agent（权重：2.5） - 数据保护
architect agent（权重：2.0） - 数据架构
reviewer agent（权重：1.5） - 处理模式
tester agent（权重：1.5） - 数据验证

ALGORITHM领域：

optimizer agent（权重：2.5） - 性能分析
architect agent（权重：2.5） - 设计模式
tester agent（权重：2.0） - 正确性测试
reviewer agent（权重：2.0） - 代码质量

Step 3: Present Decision to Agents

步骤3：向Agent呈现决策内容

Each selected agent receives:

Clear decision statement
Available options (if applicable)
Relevant context and constraints
Evaluation criteria

Decision Prompt Template:

markdown

undefined

每个选中的Agent会收到：

清晰的决策说明
可选方案（如有）
相关上下文与约束
评估标准

决策提示模板：

markdown

undefined

Decision Required: [TITLE]

待决策事项：[标题]

Domain: [SECURITY | AUTH | DATA | ALGORITHM] Your Weight: [X.X] (based on domain expertise)

领域： [SECURITY | AUTH | DATA | ALGORITHM] 你的权重： [X.X]（基于领域专业度）

Context

上下文

[Relevant background and constraints]

[相关背景与约束]

Options

可选方案

[Reject Both]: Propose alternative

[方案A]：[描述]
[方案B]：[描述]
[拒绝两者]：提出替代方案

Evaluation Criteria

评估标准

Security implications
Implementation complexity
Maintainability
Risk assessment

安全影响
实现复杂度
可维护性
风险评估

Your Vote

你的投票

Provide:

Your vote (Option 1, 2, or Reject)
Confidence level (HIGH, MEDIUM, LOW)
Key reasoning (2-3 sentences max)
Any conditions or caveats

undefined

请提供：

投票选择（方案1、2或拒绝）
置信度（HIGH、MEDIUM、LOW）
核心理由（最多2-3句话）
任何条件或注意事项

undefined

Step 4: Collect Votes

步骤4：收集投票

For each agent, collect structured vote:

yaml

agent: security
weight: 3.0
vote: Option 1
confidence: HIGH
reasoning: "Option 1 follows OWASP best practices for credential storage. Option 2 uses deprecated hashing algorithm."
conditions: ["Ensure salt length >= 16 bytes", "Use constant-time comparison"]

收集每个Agent的结构化投票：

yaml

agent: security
weight: 3.0
vote: Option 1
confidence: HIGH
reasoning: "Option 1 follows OWASP best practices for credential storage. Option 2 uses deprecated hashing algorithm."
conditions: ["Ensure salt length >= 16 bytes", "Use constant-time comparison"]

Step 5: Calculate Weighted Result

步骤5：计算加权结果

Weighted Vote Calculation:

For each option:
  weighted_score = sum(agent_weight * confidence_multiplier)

Where confidence_multiplier:
  HIGH = 1.0
  MEDIUM = 0.7
  LOW = 0.4

Example Calculation:

Agent	Weight	Vote	Confidence	Score
security	3.0	A	HIGH	3.0
architect	1.5	A	MEDIUM	1.05
reviewer	1.5	B	HIGH	1.5
tester	1.0	A	LOW	0.4

Option A Score: 3.0 + 1.05 + 0.4 = 4.45
Option B Score: 1.5
Total Weighted Votes: 5.95
Option A Percentage: 74.8% (SUPERMAJORITY)

加权投票计算方式：

针对每个选项：
  weighted_score = sum(agent_weight * confidence_multiplier)

其中置信度乘数：
  HIGH = 1.0
  MEDIUM = 0.7
  LOW = 0.4

计算示例：

Agent	权重	投票	置信度	得分
security	3.0	A	HIGH	3.0
architect	1.5	A	MEDIUM	1.05
reviewer	1.5	B	HIGH	1.5
tester	1.0	A	LOW	0.4

方案A得分：3.0 + 1.05 + 0.4 = 4.45
方案B得分：1.5
加权投票总分：5.95
方案A占比：74.8%（符合超级多数要求）

Step 6: Apply Voting Threshold

步骤6：应用投票阈值

Based on configured voting mode:

simple-majority (50%+):

Option with highest weighted score wins if > 50%
If no option > 50%, proceed to debate or reject

supermajority (66%+): (DEFAULT for security)

Winning option must have > 66% weighted votes
Provides stronger validation for high-risk decisions

unanimous (100%):

All agents must agree (rare, highest bar)
Any dissent blocks decision

根据配置的投票模式：

simple-majority（超过50%）：

加权得分最高的选项若超过50%则获胜
若无选项超过50%，则进入辩论或拒绝

supermajority（超过66%）：（安全场景默认值）

获胜选项必须获得超过66%的加权投票
为高风险决策提供更强的验证

unanimous（100%）：

所有Agent必须同意（罕见，要求最高）
任何异议都会阻止决策通过

Step 7: Report Consensus Result

步骤7：报告共识结果

If Consensus Reached:

markdown

undefined

达成共识时：

markdown

undefined

Consensus Voting Result

共识投票结果

Decision: [Selected Option] Domain: SECURITY Threshold: Supermajority (66%+) Result: PASSED (74.8%)

决策： [选中方案] 领域： SECURITY 阈值： 超级多数（66%+） 结果： 通过（74.8%）

Vote Summary

投票汇总

Agent	Weight	Vote	Confidence
security	3.0	A	HIGH
architect	1.5	A	MEDIUM
reviewer	1.5	B	HIGH
tester	1.0	A	LOW

Agent	权重	投票	置信度
security	3.0	A	HIGH
architect	1.5	A	MEDIUM
reviewer	1.5	B	HIGH
tester	1.0	A	LOW

Key Reasoning (from highest-weighted agents)

核心理由（来自最高权重Agent）

security (3.0): "Option 1 follows OWASP best practices..."

security (3.0)： "方案1遵循OWASP凭证存储最佳实践..."

Conditions/Caveats

条件/注意事项

Ensure salt length >= 16 bytes
Use constant-time comparison

确保盐长度 >= 16字节
使用恒定时间比较

Dissenting View

异议观点

reviewer (1.5): "Option B has simpler implementation..."


**If No Consensus:**

```markdown

reviewer (1.5)： "方案B实现更简单..."


**未达成共识时：**

```markdown

Consensus Voting Result

共识投票结果

Decision: NO CONSENSUS Domain: AUTH Threshold: Supermajority (66%+) Result: FAILED (52.3%)

决策： 未达成共识 领域： AUTH 阈值： 超级多数（66%+） 结果： 未通过（52.3%）

Recommendation

建议

Escalate to
```
/amplihack:debate
```
for structured trade-off analysis
OR: Gather more information and re-vote
OR: Accept simple majority with documented risk

undefined

升级至
```
/amplihack:debate
```
进行结构化权衡分析
或：收集更多信息后重新投票
或：接受简单多数并记录风险

undefined

Step 8: Record Decision

步骤8：记录决策

Log voting result to session decisions
Document vote reasoning for future reference

将投票结果记录到会话决策中
记录投票理由，以备未来参考

Trade-Offs

权衡

Benefits:

Structured decision-making for high-risk domains
Domain expertise appropriately weighted
Clear audit trail with reasoning
Faster than full debate for binary/discrete choices

Costs:

Less nuanced than debate (no synthesis)
Requires clear options (not generative)
Weight calibration needs real data
May miss creative alternatives

Use When: Decision is discrete, domain expertise matters, audit trail needed

优势：

为高风险领域提供结构化决策机制
领域专业度得到合理加权
包含理由的清晰审计追踪
比完整辩论更适合二元/离散选择

成本：

不如辩论细致（无合成过程）
需要明确的可选方案（不适合生成式场景）
权重校准需要真实数据
可能错过创新性替代方案

适用场景： 决策为离散型、领域专业度重要、需要审计追踪的情况

Examples

示例

Example 1: Password Hashing Implementation

示例1：密码哈希实现选择

Context: Choosing between bcrypt, argon2id, and PBKDF2 for new auth system

Domain Detection: AUTH (password, hash)

Agents Selected:

security (2.5), architect (2.0), reviewer (1.5), tester (1.5)

Votes:

security: argon2id (HIGH) - "OWASP current recommendation, memory-hard"
architect: argon2id (MEDIUM) - "Good library support, modern design"
reviewer: bcrypt (HIGH) - "More battle-tested in production"
tester: argon2id (MEDIUM) - "Easier to test with configurable params"

Result: argon2id wins with 68.2% (supermajority passed)

场景： 为新认证系统选择bcrypt、argon2id或PBKDF2

领域检测： AUTH（password、hash）

选中Agent：

security (2.5)、architect (2.0)、reviewer (1.5)、tester (1.5)

投票：

security: argon2id（HIGH）- "OWASP当前推荐的内存密集型算法"
architect: argon2id（MEDIUM）- "库支持良好，设计现代"
reviewer: bcrypt（HIGH）- "生产环境中经过更多实战检验"
tester: argon2id（MEDIUM）- "参数配置更易于测试"

结果： argon2id以68.2%的支持率通过（符合超级多数要求）

Example 2: API Rate Limiting Approach

示例2：API限流方案选择

Context: Token bucket vs sliding window vs fixed window

Domain Detection: SECURITY (rate limit, protection)

Agents Selected:

security (3.0), optimizer (1.5), architect (1.5), reviewer (1.5)

Votes:

security: token bucket (HIGH) - "Best protection against burst attacks"
optimizer: sliding window (MEDIUM) - "Better resource utilization"
architect: token bucket (MEDIUM) - "Industry standard, well-understood"
reviewer: sliding window (LOW) - "Simpler implementation"

Result: token bucket wins with 69.4% (supermajority passed)

场景： 令牌桶、滑动窗口还是固定窗口

领域检测： SECURITY（rate limit、protection）

选中Agent：

security (3.0)、optimizer (1.5)、architect (1.5)、reviewer (1.5)

投票：

security: 令牌桶（HIGH）- "对突发攻击的防护效果最佳"
optimizer: 滑动窗口（MEDIUM）- "资源利用率更高"
architect: 令牌桶（MEDIUM）- "行业标准，易于理解"
reviewer: 滑动窗口（LOW）- "实现更简单"

结果： 令牌桶以69.4%的支持率通过（符合超级多数要求）

Example 3: No Consensus Scenario

示例3：未达成共识场景

Context: Microservices vs monolith for new feature

Domain Detection: ALGORITHM (general architecture)

Agents Selected:

architect (2.5), optimizer (2.5), reviewer (2.0), tester (2.0)

Votes:

architect: microservices (MEDIUM) - "Better long-term scalability"
optimizer: monolith (HIGH) - "Simpler operations, less overhead"
reviewer: monolith (MEDIUM) - "Easier to maintain initially"
tester: microservices (LOW) - "Harder to test but more isolated"

Result: No consensus (52.1%) - Escalate to

/amplihack:debate

场景： 新功能采用微服务还是单体架构

领域检测： ALGORITHM（通用架构）

选中Agent：

architect (2.5)、optimizer (2.5)、reviewer (2.0)、tester (2.0)

投票：

architect: 微服务（MEDIUM）- "长期扩展性更好"
optimizer: 单体（HIGH）- "运维更简单，开销更小"
reviewer: 单体（MEDIUM）- "初期维护更轻松"
tester: 微服务（LOW）- "测试难度大但隔离性更好"

结果： 未达成共识（52.1%）- 升级至

/amplihack:debate

Integration with Other Workflows

与其他工作流集成

Handoff to Debate

移交至辩论工作流

When voting fails to reach consensus:

Document vote results and reasoning
Invoke
```
Skill(debate-workflow)
```
with vote context
Use voting insights to frame debate perspectives

当投票无法达成共识时：

记录投票结果与理由
调用
```
Skill(debate-workflow)
```
并传入投票上下文
利用投票洞察构建辩论视角

After N-Version Implementation

N版本实现后集成

Use consensus voting to select between N-version implementations:

N-version generates 3+ implementations
Consensus voting selects winner
Domain experts have weighted influence on selection

使用共识投票从N版本实现中选择最优方案：

N版本生成3个以上实现方案
共识投票选出获胜方案
领域专家的加权意见影响选择结果

Philosophy Alignment

理念对齐

This skill enforces:

Evidence-Based Decisions: Votes require reasoning
Domain Expertise: Weights reflect competence
Transparent Trade-offs: Dissent documented
Audit Trail: Full voting record preserved
Appropriate Rigor: Auto-triggers for high-risk domains

本Skill遵循以下原则：

基于证据的决策： 投票需附带理由
领域专业度： 权重反映能力水平
透明权衡： 异议会被记录
审计追踪： 完整投票记录被保留
适度严谨： 高风险领域自动触发