AgentSkillsCN

team-shinchan:debate

通过专业代理间的辩论,寻找最优解决方案。适用于“辩论”、“利弊分析”、“收集意见”等需求。

SKILL.md
--- frontmatter
name: team-shinchan:debate
description: Specialized agents debate to find optimal solutions. Used for "debate", "pros and cons", "gather opinions" requests.
user-invocable: true

EXECUTE IMMEDIATELY - Delegate to Midori

Invoke Midori using the Task tool to proceed with Debate.


🔔 Auto-Trigger Conditions

Shinnosuke starts Debate immediately when detecting the following situations:

SituationAuto-DebateExample
2+ implementation approaches existREST vs GraphQL, Monolith vs Microservices
Architecture changeDB schema redesign, layer structure change
Breaking existing patternsProposing different approach from existing conventions
Performance vs Readability tradeoffOptimization vs Maintainability
Security-sensitive decisionsAuthentication method, data encryption approach
Simple CRUDSimple CRUD endpoints
Clear bug fixObvious bug fix
User explicitly decidedUser has already decided

Auto-Trigger Behavior

  1. Immediately announce Debate start

    code
    ⚠️ Design decision needed: [detected situation]
    → Starting Debate automatically
    
  2. Execute same process as manual invocation

    • Proceed with Steps 1-3 below
    • Difference: Add background explanation since user didn't explicitly invoke
  3. Record decision in REQUESTS.md

    • Stage 1: Add decision to requirements
    • Stage 2+: Record in corresponding Phase in PROGRESS.md

Step 1: Invoke Midori

typescript
Task(
  subagent_type="team-shinchan:midori",
  model="opus",
  prompt="Please proceed with Debate.

## Topic
{discussion topic}

## Panel
{panel list}

## Procedure
1. Announce Debate start
2. Collect panel opinions (parallel Tasks)
3. Output opinions in real-time
4. Hiroshi derives consensus
5. Report final decision"
)

Step 2: Deliver Results to User

When receiving results from Midori, deliver to user in the following format:

code
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
💭 Debate Results
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📋 Topic: {topic}

🎤 Expert Opinions:
- [Hiroshi]: {opinion summary}
- [Nene]: {opinion summary}

✅ Recommended Decision: {Midori's conclusion}
📝 Rationale: {reasoning}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Step 3: Confirm User Opinion

After delivering results, ask the user:

"Do you agree with the recommended decision? If you have other opinions or additional considerations, please let me know."

Step 4: Final Decision

  • If user agrees: Document decision and proceed
  • If user disagrees: Revise decision reflecting concerns
  • Never proceed without user confirmation

Panel Selection Criteria

TopicPanel
UI/FrontendAichan, Hiroshi
API/BackendBunta, Hiroshi
DevOps/InfraMasao, Hiroshi
ArchitectureHiroshi, Nene, Misae

📖 Auto-Trigger Examples

Example 1: Detecting 2+ Implementation Approaches

code
[Shinnosuke analyzing...]
Detected: JWT and Session both possible for authentication implementation

⚠️ Design decision needed: Choose authentication method (JWT vs Session)
→ Starting Debate automatically

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
💭 Debate Started (auto)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📋 Topic: Authentication method selection
👥 Panel: Hiroshi, Bunta
🎯 Goal: Choose optimal between JWT and Session

[Regular Debate process follows...]

Example 2: Detecting Architecture Change

code
[Bo proposes...]
"Instead of adding roles field to User table, I propose creating separate Role table."

[Shinnosuke detects]
Detected: DB schema change → Architecture impact review needed

⚠️ Design decision needed: Role management approach (single table vs normalization)
→ Starting Debate automatically

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
💭 Debate Started (auto)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📋 Topic: Role management DB design
👥 Panel: Hiroshi, Bunta, Nene
🎯 Goal: Determine optimal schema considering scalability and maintainability

Example 3: Performance vs Readability Tradeoff

code
[Action Kamen reviewing...]
"Current code has good readability but N+1 query issue exists. Optimization increases complexity."

[Shinnosuke detects]
Detected: Performance optimization vs code readability tradeoff

⚠️ Design decision needed: Determine query optimization level
→ Starting Debate automatically

⚙️ Shinnosuke's Auto-Detection Logic

Shinnosuke detects the following signals:

SignalDetection Method
2+ approaches mentionedExpressions like "A or B", "vs", "method1/method2"
Architecture keywords"schema change", "layer", "structure", "architecture"
Pattern violationsAction Kamen warns "differs from existing pattern"
Tradeoff mentions"but", "however", "trade-off", "at the cost of"
Security keywords"auth", "security", "encryption", "permission"

After auto-detection, immediately:

  1. Announce situation
  2. Start Debate (same as above process)
  3. Document decision