agileflow 3.4.3 → 4.0.0-alpha.2

package/src/core/commands/multi-expert.md

@@ -1,447 +0,0 @@
----
-description: Deploy multiple domain experts on the same problem for higher confidence
-phase: planning
-argument-hint: "<question>"
-compact_context:
-  priority: high
-  preserve_rules:
-    - "ACTIVE COMMAND: /agileflow:multi-expert - Parallel expert orchestration"
-    - "CRITICAL: Deploy ALL experts in SINGLE message with multiple Task tool calls (not sequential)"
-    - "MUST wait for all experts via TaskOutput with block=true before synthesis"
-    - "MUST analyze agreement: 3+ experts agree = HIGH confidence, 2 agree = MEDIUM, 1 = unique insight"
-    - "MUST detect domain keywords from user question to select 3-5 experts"
-    - "Synthesis format: Key Findings (agreement) | Unique Insights (single expert) | Disagreements (needs review)"
-  state_fields:
-    - selected_experts
-    - user_question
-    - expert_results
-    - confidence_scores
----
-
-# /agileflow-multi-expert
-
-Deploy multiple Agent Experts on the same problem. Each expert validates independently, then results are synthesized for higher confidence answers.
-
-<!-- COMPACT_SUMMARY_START -->
-## Compact Summary
-
-**Command**: `/agileflow:multi-expert` - Deploy 3-5 domain experts in parallel for high-confidence analysis
-
-**Quick Usage**:
-```
-/agileflow:multi-expert Is our authentication implementation secure?
-```
-
-**What It Does**: Analyze question for domains → Select 3-5 experts → Deploy in PARALLEL → Collect results → Synthesize with confidence scoring
-
-**Experts Available**: Security, API, Testing, Database, Performance, CI/CD, DevOps, Accessibility, Architecture
-
-**Critical Rules**:
-- 🚨 Deploy ALL experts in ONE message with multiple Task calls (not sequential)
-- 🚨 Wait for all results before synthesis (use TaskOutput with block=true)
-- 🚨 Confidence: HIGH (3+ agree) | MEDIUM (2 agree) | UNIQUE INSIGHT (1 expert with evidence)
-- Detect domain keywords to select 3-5 experts (max 5 to avoid overhead)
-
-**Output Format**: Key Findings (high confidence) | Unique Insights (single expert) | Disagreements (needs review) | Recommended Actions
-
-**Tool Usage Examples**:
-
-**Task** (deploy expert in parallel):
-```xml
-<invoke name="Task">
-<parameter name="description">[Domain] analysis of: {question}</parameter>
-<parameter name="prompt">EXPERTISE FIRST: Read packages/cli/src/core/experts/{domain}/expertise.yaml
-
-QUESTION: {user_question}
-
-Analyze from your {domain} perspective:
-1. What do you observe in your domain?
-2. What concerns or issues do you see?
-3. What recommendations do you have?
-4. Confidence level (High/Medium/Low) and why?
-
-Be specific with file paths and code references.</parameter>
-<parameter name="subagent_type">agileflow-{domain}</parameter>
-<parameter name="run_in_background">true</parameter>
-</invoke>
-```
-
-**TaskOutput** (collect all results):
-```xml
-<invoke name="TaskOutput">
-<parameter name="task_id">{expert_task_id}</parameter>
-<parameter name="block">true</parameter>
-</invoke>
-```
-
-**Domain Detection Keywords**:
-| Domain | Keywords |
-|--------|----------|
-| Security | auth, JWT, OAuth, vulnerability, XSS, CSRF, secure, encrypt |
-| API | endpoint, REST, GraphQL, route, controller, backend |
-| Testing | test, spec, coverage, mock, assertion |
-| Database | schema, table, query, migration, SQL, model |
-| Performance | optimize, cache, latency, profiling, slow |
-
-<!-- COMPACT_SUMMARY_END -->
-
-## When to Use
-
-- **Complex cross-domain questions** - "How does user authentication flow from login to API?"
-- **Architecture analysis** - "What are the security implications of this change?"
-- **Code review with multiple perspectives** - "Review this PR from different angles"
-- **Debugging hard problems** - "Why is this test flaky?"
-- **Best practice validation** - "Is our approach correct?"
-
-## How It Works
-
-```
-┌─────────────────────────────────────────────────────────────┐
-│                    USER QUESTION                             │
-└─────────────────────────────────────────────────────────────┘
-                            │
-                            ▼
-┌─────────────────────────────────────────────────────────────┐
-│              MULTI-EXPERT ORCHESTRATOR                       │
-│  1. Analyze question for relevant domains                    │
-│  2. Select 3-5 experts based on domain detection             │
-│  3. Deploy experts IN PARALLEL                               │
-│  4. Collect and synthesize results                           │
-│  5. Return high-confidence answer                            │
-└─────────────────────────────────────────────────────────────┘
-                            │
-            ┌───────────────┼───────────────┐
-            ▼               ▼               ▼
-    ┌───────────┐   ┌───────────┐   ┌───────────┐
-    │  Expert 1 │   │  Expert 2 │   │  Expert 3 │
-    │ (domain A)│   │ (domain B)│   │ (domain C)│
-    │           │   │           │   │           │
-    │ 1. Load   │   │ 1. Load   │   │ 1. Load   │
-    │  expertise│   │  expertise│   │  expertise│
-    │ 2. Validate│  │ 2. Validate│  │ 2. Validate│
-    │ 3. Analyze │  │ 3. Analyze │  │ 3. Analyze │
-    │ 4. Report  │  │ 4. Report  │  │ 4. Report  │
-    └───────────┘   └───────────┘   └───────────┘
-            │               │               │
-            └───────────────┼───────────────┘
-                            ▼
-┌─────────────────────────────────────────────────────────────┐
-│                   RESULT SYNTHESIS                           │
-│  • Find areas of agreement (high confidence)                 │
-│  • Identify unique insights from each expert                 │
-│  • Flag disagreements for human review                       │
-│  • Combine into comprehensive answer                         │
-└─────────────────────────────────────────────────────────────┘
-                            │
-                            ▼
-┌─────────────────────────────────────────────────────────────┐
-│                   FINAL ANSWER                               │
-│  📊 Confidence: High/Medium/Low                              │
-│  🎯 Key Findings (agreed by multiple experts)                │
-│  💡 Unique Insights (from individual experts)                │
-│  ⚠️ Disagreements (needs human review)                       │
-└─────────────────────────────────────────────────────────────┘
-```
-
-## Prompt
-
-ROLE: Multi-Expert Orchestrator
-
-You coordinate multiple Agent Experts to analyze the same problem from different perspectives, then synthesize their findings into a high-confidence answer.
-
-### STEP 1: DOMAIN ANALYSIS
-
-Analyze the user's question/task to identify relevant domains.
-
-**Domain Detection Keywords**:
-
-| Domain | Keywords | Expert |
-|--------|----------|--------|
-| Database | schema, table, query, migration, SQL, model | agileflow-database |
-| API | endpoint, REST, GraphQL, route, controller, backend | agileflow-api |
-| UI | component, frontend, button, form, style, CSS | agileflow-ui |
-| Testing | test, spec, coverage, mock, assertion | agileflow-testing |
-| Security | auth, JWT, OAuth, vulnerability, XSS, CSRF | agileflow-security |
-| Performance | optimize, cache, latency, profiling, slow | agileflow-performance |
-| CI/CD | workflow, pipeline, build, deploy, GitHub Actions | agileflow-ci |
-| DevOps | infrastructure, Docker, Kubernetes, env | agileflow-devops |
-| Accessibility | ARIA, a11y, screen reader, WCAG | agileflow-accessibility |
-| Architecture | design, pattern, structure, decision | agileflow-adr-writer |
-
-**Selection Rules**:
-- Select 3-5 experts maximum (balance coverage vs. overhead)
-- Always include the PRIMARY domain (most relevant)
-- Add SUPPORTING domains that could provide unique insights
-- For cross-cutting concerns, include: security, testing, performance
-
-### STEP 2: DEPLOY EXPERTS IN PARALLEL
-
-Use the Task tool to spawn multiple experts simultaneously.
-
-**CRITICAL**: Deploy ALL experts in a SINGLE message with multiple Task tool calls.
-
-```
-Task(
-  description: "[Domain 1] analysis",
-  prompt: "EXPERTISE FIRST: Read packages/cli/src/core/experts/{domain1}/expertise.yaml
-
-QUESTION: {user_question}
-
-Analyze this from your {domain1} perspective:
-1. What do you observe in your domain?
-2. What concerns or issues do you see?
-3. What recommendations do you have?
-4. Confidence level (High/Medium/Low) and why?
-
-Be specific with file paths and code references.",
-  subagent_type: "agileflow-{domain1}",
-  run_in_background: true
-)
-
-Task(
-  description: "[Domain 2] analysis",
-  prompt: "...",
-  subagent_type: "agileflow-{domain2}",
-  run_in_background: true
-)
-
-Task(
-  description: "[Domain 3] analysis",
-  prompt: "...",
-  subagent_type: "agileflow-{domain3}",
-  run_in_background: true
-)
-```
-
-### STEP 3: COLLECT RESULTS
-
-Wait for all experts to complete using TaskOutput.
-
-```
-TaskOutput(task_id: "...", block: true)
-```
-
-### STEP 4: SYNTHESIZE RESULTS
-
-Analyze all expert responses and synthesize:
-
-**Agreement Analysis**:
-- What do multiple experts agree on? → HIGH CONFIDENCE
-- What does only one expert mention? → UNIQUE INSIGHT
-- Where do experts disagree? → NEEDS REVIEW
-
-**Synthesis Template**:
-
-```markdown
-## Multi-Expert Analysis Results
-
-**Question**: {original_question}
-**Experts Consulted**: {list of experts}
-**Overall Confidence**: High/Medium/Low
-
-### 🎯 Key Findings (High Confidence)
-*Areas where multiple experts agree*
-
-1. [Finding] - Agreed by: {expert1}, {expert2}
-2. [Finding] - Agreed by: {expert1}, {expert3}
-
-### 💡 Unique Insights
-*Valuable perspectives from individual experts*
-
-- **{Expert1}**: [Unique insight from their domain]
-- **{Expert2}**: [Unique insight from their domain]
-
-### ⚠️ Areas of Disagreement
-*Requires human review to resolve*
-
-- {Expert1} says X, but {Expert2} says Y
-  - Context: [why they might disagree]
-  - Recommendation: [how to resolve]
-
-### 📋 Recommended Actions
-
-1. [Action] - Priority: High - Source: {expert}
-2. [Action] - Priority: Medium - Source: {expert}
-
-### 🔍 Files Referenced
-
-- `path/to/file.ts` - Mentioned by: {experts}
-- `path/to/other.ts` - Mentioned by: {experts}
-```
-
-### CONFIDENCE SCORING
-
-**High Confidence** (report with conviction):
-- 3+ experts agree on the finding
-- Evidence is cited with specific file paths
-- No contradicting opinions
-
-**Medium Confidence** (report with caveats):
-- 2 experts agree
-- Some evidence provided
-- Minor disagreements exist
-
-**Low Confidence** (flag for human review):
-- Only 1 expert mentions it
-- No specific evidence
-- Major disagreements between experts
-
-### EXAMPLE ORCHESTRATION
-
-**User Question**: "Is our authentication implementation secure?"
-
-**Step 1 - Domain Analysis**:
-- Primary: Security (auth, secure)
-- Supporting: API (implementation likely in API layer)
-- Supporting: Testing (should verify test coverage)
-- Supporting: Database (if storing credentials)
-
-**Step 2 - Deploy Experts**:
-```
-Task(subagent_type: "agileflow-security", prompt: "Analyze auth security...", run_in_background: true)
-Task(subagent_type: "agileflow-api", prompt: "Analyze auth API implementation...", run_in_background: true)
-Task(subagent_type: "agileflow-testing", prompt: "Analyze auth test coverage...", run_in_background: true)
-Task(subagent_type: "agileflow-database", prompt: "Analyze credential storage...", run_in_background: true)
-```
-
-**Step 3 - Collect & Synthesize**:
-- Security expert: "JWT tokens don't expire, HIGH RISK"
-- API expert: "Auth middleware is applied inconsistently"
-- Testing expert: "Only 40% coverage on auth routes"
-- Database expert: "Passwords are hashed with bcrypt, good"
-
-**Step 4 - Final Answer**:
-```
-## Multi-Expert Analysis: Authentication Security
-
-**Overall Confidence**: High (4 experts, strong agreement)
-
-### 🎯 Key Findings
-1. **JWT tokens never expire** - CRITICAL - Agreed by: Security, API
-2. **Inconsistent middleware application** - HIGH - Agreed by: Security, API
-3. **Low test coverage (40%)** - MEDIUM - Testing
-
-### 💡 Unique Insights
-- **Database**: Password hashing is correctly implemented (bcrypt)
-
-### 📋 Recommended Actions
-1. Add JWT expiration (Priority: CRITICAL)
-2. Audit all routes for auth middleware (Priority: HIGH)
-3. Increase auth test coverage to 80% (Priority: MEDIUM)
-```
-
-### FIRST MESSAGE
-
-When invoked, immediately:
-
-1. **Acknowledge the question**
-2. **Identify relevant domains** (show your analysis)
-3. **List selected experts** (explain why each)
-4. **Deploy experts in parallel**
-5. **Wait and synthesize**
-6. **Present comprehensive answer**
-
-Example first message:
-```
-Analyzing your question for relevant domains...
-
-**Detected Domains**:
-- Primary: Security (keywords: "secure", "auth")
-- Supporting: API (auth likely in API layer)
-- Supporting: Testing (verify coverage)
-
-**Deploying 3 experts in parallel**:
-1. 🔒 Security Expert - Vulnerability analysis
-2. 🔌 API Expert - Implementation review
-3. 🧪 Testing Expert - Coverage analysis
-
-Launching experts now...
-```
-
-### ARGUMENTS
-
-{{argument}}
-
----
-
-## Expected Output
-
-### Success - Multi-Expert Analysis
-
-```
-🧠 Multi-Expert Analysis
-══════════════════════════════════════════════════════════════
-
-Question: Should we add caching to the user API?
-
-Deploying 3 experts...
-✓ Performance Expert launched
-✓ Security Expert launched
-✓ Architecture Expert launched
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-⚡ PERFORMANCE EXPERT
-"Yes, caching would reduce DB load by ~60% based on
-read patterns. Recommend Redis with 5-minute TTL."
-Confidence: 90%
-
-🔒 SECURITY EXPERT
-"Ensure cache invalidation on sensitive updates.
-Use per-user cache keys to prevent data leakage."
-Confidence: 85%
-
-🏛️ ARCHITECTURE EXPERT
-"Start with single-node Redis. Cluster only if
-needed for HA. Consider cache-aside pattern."
-Confidence: 88%
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-📊 Consensus: YES (88% confidence)
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-Synthesized Recommendation:
-Add Redis caching with cache-aside pattern,
-5-minute TTL, per-user keys, and proper
-invalidation on user updates.
-
-Document this decision? /agileflow:adr
-```
-
-### Success - Experts Disagree
-
-```
-🧠 Multi-Expert Analysis
-══════════════════════════════════════════════════════════════
-
-⚠️ Expert Disagreement Detected
-
-Experts split on approach:
-- Performance: Recommends Redis (90%)
-- Cost Expert: Recommends in-memory (85%)
-- Ops Expert: Recommends managed service (80%)
-
-Run debate mode for resolution?
-/agileflow:council --mode debate
-```
-
-### Error - Insufficient Context
-
-```
-❌ Error: Question too vague for expert analysis
-
-"Is this good?" - What is "this"?
-
-Provide specific context:
-/agileflow:multi-expert "Should we use Redis or Memcached for session storage given our 10K DAU scale?"
-```
-
----
-
-## Related Commands
-
-- `/agileflow:review` - AI-powered code review
-- `/agileflow:diagnose` - System health diagnostics
-- `/agileflow:impact` - Analyze change impact
-- `/agileflow:babysit` - Mentor workflow
-- `/agileflow:rpi` - Research-Plan-Implement workflow