🇺🇸English

Group Collaboration Skill

Overview

This skill provides guidelines, patterns, and best practices for effective collaboration between the four agent groups in the four-tier architecture. It covers communication protocols, knowledge transfer strategies, feedback mechanisms, and coordination patterns that enable autonomous learning and continuous improvement across groups.

When to Apply This Skill

Use this skill when:

• Implementing inter-group communication between any two groups
• Designing handoff protocols between analysis, decision, execution, and validation phases
• Setting up feedback loops for continuous improvement
• Sharing knowledge and patterns across groups
• Coordinating multi-group workflows
• Troubleshooting collaboration issues between groups
• Optimizing group performance through better coordination

Required for:

• All agents in four-tier architecture (Groups 1, 2, 3, 4)
• Orchestrator coordination logic
• Cross-group pattern learning
• Workflow optimization

Four-Tier Architecture Recap

Group 1: Strategic Analysis & Intelligence (The "Brain")

• Role : Analyze and recommend
• Output : Recommendations with confidence scores
• Key Agents : code-analyzer, security-auditor, smart-recommender

Group 2: Decision Making & Planning (The "Council")

• Role : Evaluate and decide
• Output : Execution plans with priorities
• Key Agents : strategic-planner, preference-coordinator

Group 3: Execution & Implementation (The "Hand")

• Role : Execute decisions
• Output : Execution results with metrics
• Key Agents : quality-controller, test-engineer, documentation-generator

Group 4: Validation & Optimization (The "Guardian")

• Role : Validate and optimize
• Output : Validation results and feedback
• Key Agents : post-execution-validator, performance-optimizer, continuous-improvement

Communication Patterns

Pattern 1: Analysis → Decision (Group 1 → Group 2)

Purpose : Transfer analysis findings and recommendations to decision-makers

Structure :

from lib.group_collaboration_system import record_communication

record_communication(
    from_agent="code-analyzer",  # Group 1
    to_agent="strategic-planner",  # Group 2
    task_id=task_id,
    communication_type="recommendation",
    message="Code analysis complete with 5 recommendations",
    data={
        "quality_score": 72,
        "recommendations": [
            {
                "type": "refactoring",
                "priority": "high",
                "confidence": 0.92,  # High confidence
                "description": "Extract login method complexity",
                "rationale": "Cyclomatic complexity 15, threshold 10",
                "estimated_effort_hours": 2.5,
                "expected_impact": "high",
                "files_affected": ["src/auth.py"]
            }
        ],
        "patterns_detected": ["token_auth", "validation_duplication"],
        "metrics": {
            "complexity_avg": 8.5,
            "duplication_rate": 0.12,
            "test_coverage": 0.78
        }
    }
)

Best Practices :

• Always include confidence scores (0.0-1.0)
• Provide rationale for each recommendation
• Include estimated effort and expected impact
• Attach relevant metrics and context
• Reference detected patterns
• List affected files

Anti-Patterns to Avoid :

• ❌ Recommendations without confidence scores
• ❌ Missing rationale or context
• ❌ Vague impact estimates ("it will be better")
• ❌ No prioritization
• ❌ Execution commands (that's Group 3's job)

Pattern 2: Decision → Execution (Group 2 → Group 3)

Purpose : Communicate execution plan with priorities and user preferences

Structure :

record_communication(
    from_agent="strategic-planner",  # Group 2
    to_agent="quality-controller",  # Group 3
    task_id=task_id,
    communication_type="execution_plan",
    message="Execute quality improvement plan with 3 priorities",
    data={
        "decision_rationale": "High-priority refactoring based on user preferences",
        "execution_plan": {
            "quality_targets": {
                "tests": 80,
                "standards": 90,
                "documentation": 70
            },
            "priority_order": [
                "fix_failing_tests",  # Highest priority
                "apply_code_standards",
                "add_missing_docs"
            ],
            "approach": "incremental",  # or "comprehensive"
            "risk_tolerance": "low"  # User preference
        },
        "user_preferences": {
            "auto_fix_threshold": 0.9,
            "coding_style": "concise",
            "comment_level": "moderate",
            "documentation_level": "standard"
        },
        "constraints": {
            "max_iterations": 3,
            "time_budget_minutes": 15,
            "files_in_scope": ["src/auth.py", "src/utils.py"]
        },
        "decision_confidence": 0.88
    }
)

Best Practices :

• Include clear execution plan with priorities
• Apply user preferences to the plan
• Set realistic constraints (time, iterations)
• Provide decision rationale
• Specify risk tolerance
• Define success criteria

Anti-Patterns to Avoid :

• ❌ Plans without priorities
• ❌ Missing user preferences
• ❌ Unrealistic constraints
• ❌ No success criteria
• ❌ Ambiguous instructions

Pattern 3: Execution → Validation (Group 3 → Group 4)

Purpose : Send execution results for validation and quality assessment

Structure :

record_communication(
    from_agent="quality-controller",  # Group 3
    to_agent="post-execution-validator",  # Group 4
    task_id=task_id,
    communication_type="execution_result",
    message="Quality improvement complete: 68 → 84",
    data={
        "metrics_before": {
            "quality_score": 68,
            "tests_passing": 45,
            "standards_violations": 23,
            "doc_coverage": 0.60
        },
        "metrics_after": {
            "quality_score": 84,
            "tests_passing": 50,
            "standards_violations": 2,
            "doc_coverage": 0.75
        },
        "changes_made": {
            "tests_fixed": 5,
            "standards_violations_fixed": 21,
            "docs_generated": 10
        },
        "files_modified": ["src/auth.py", "tests/test_auth.py"],
        "auto_corrections_applied": 30,
        "manual_review_needed": [],
        "iterations_used": 2,
        "execution_time_seconds": 145,
        "component_scores": {
            "tests": 28,
            "standards": 22,
            "documentation": 16,
            "patterns": 13,
            "code_metrics": 5
        },
        "issues_encountered": []
    }
)

Best Practices :

• Show before/after metrics clearly
• List all changes made
• Include execution statistics
• Report any issues encountered
• Specify files modified
• Break down component scores

Anti-Patterns to Avoid :

• ❌ Only showing final metrics without before state
• ❌ Missing execution time and iterations
• ❌ No breakdown of what was changed
• ❌ Hiding issues or failures
• ❌ Incomplete component scoring

Pattern 4: Validation → Analysis (Group 4 → Group 1)

Purpose : Provide feedback on recommendation effectiveness for learning

Structure :

from lib.agent_feedback_system import add_feedback

add_feedback(
    from_agent="post-execution-validator",  # Group 4
    to_agent="code-analyzer",  # Group 1
    task_id=task_id,
    feedback_type="success",  # or "improvement", "warning", "error"
    message="Recommendations were highly effective",
    details={
        "recommendations_followed": 3,
        "recommendations_effective": 3,
        "quality_improvement": 16,  # points improved
        "execution_smooth": True,
        "user_satisfaction": "high",
        "suggestions_for_improvement": []
    },
    impact="quality_score +16, all recommendations effective"
)

Best Practices :

• Specific feedback on recommendation effectiveness
• Quantify impact (quality score improvement)
• Note which recommendations worked best
• Suggest improvements for future
• Track user satisfaction

Anti-Patterns to Avoid :

• ❌ Vague feedback ("it was good")
• ❌ No quantified impact
• ❌ Only negative feedback without suggestions
• ❌ Missing context about what worked
• ❌ Not closing the feedback loop

Knowledge Transfer Strategies

Strategy 1: Pattern Propagation

When to Use : Share successful patterns across groups

from lib.inter_group_knowledge_transfer import add_knowledge

add_knowledge(
    source_group=1,  # Group 1 discovered this
    knowledge_type="pattern",
    title="Modular Authentication Pattern",
    description="Breaking auth logic into validate(), authenticate(), authorize() improves testability and maintainability",
    context={
        "applies_to": ["authentication", "authorization", "security"],
        "languages": ["python", "typescript"],
        "frameworks": ["flask", "fastapi"]
    },
    evidence={
        "quality_score_improvement": 12,
        "test_coverage_improvement": 0.15,
        "reuse_count": 5,
        "success_rate": 0.92
    }
)

Strategy 2: Anti-Pattern Sharing

When to Use : Share what NOT to do based on failures

add_knowledge(
    source_group=3,  # Group 3 encountered this during execution
    knowledge_type="anti_pattern",
    title="Avoid Nested Ternary Operators",
    description="Nested ternary operators reduce readability and increase cognitive complexity significantly",
    context={
        "applies_to": ["code_quality", "readability"],
        "severity": "medium"
    },
    evidence={
        "complexity_increase": 8,  # Cyclomatic complexity
        "maintenance_issues": 3,
        "refactoring_time_hours": 1.5
    }
)

Strategy 3: Best Practice Sharing

When to Use : Share techniques that consistently work well

add_knowledge(
    source_group=4,  # Group 4 validated this across tasks
    knowledge_type="best_practice",
    title="Test Fixtures with CASCADE for PostgreSQL",
    description="Always use CASCADE in test fixture teardown to avoid foreign key constraint errors",
    context={
        "applies_to": ["testing", "database"],
        "frameworks": ["pytest"],
        "databases": ["postgresql"]
    },
    evidence={
        "success_rate": 1.0,
        "fixes_applied": 15,
        "issues_prevented": 30
    }
)

Strategy 4: Optimization Tip Sharing

When to Use : Share performance improvements

add_knowledge(
    source_group=4,  # Group 4 performance-optimizer discovered this
    knowledge_type="optimization",
    title="Batch Database Queries in Loops",
    description="Replace N+1 query patterns with batch queries using IN clause or JOINs",
    context={
        "applies_to": ["performance", "database"],
        "orm": ["sqlalchemy", "sequelize"]
    },
    evidence={
        "performance_improvement": "80%",  # 5x faster
        "query_reduction": 0.95,  # 95% fewer queries
        "cases_improved": 8
    }
)

Feedback Loop Best Practices

1. Timely Feedback

Principle : Provide feedback immediately after validation, not days later

# ✅ GOOD: Immediate feedback
validate_results()
send_feedback_to_group_1()
send_feedback_to_group_3()

# ❌ BAD: Delayed feedback loses context
validate_results()
# ... days later ...
send_feedback()  # Context is lost

2. Actionable Feedback

Principle : Feedback must be specific and actionable, not vague

# ✅ GOOD: Specific and actionable
add_feedback(
    message="Recommendation confidence was too high (0.92) for untested pattern. Consider 0.75-0.85 for new patterns",
    suggestions=["Add confidence penalty for untested patterns", "Increase confidence gradually with reuse"]
)

# ❌ BAD: Vague
add_feedback(
    message="Confidence was wrong",
    suggestions=[]
)

3. Balanced Feedback

Principle : Highlight successes and areas for improvement

# ✅ GOOD: Balanced
add_feedback(
    positive=[
        "Priority ranking was excellent - high priority items were truly critical",
        "User preference integration worked perfectly"
    ],
    improvements=[
        "Estimated effort was 40% too low - consider adjusting effort formula",
        "Could benefit from more error handling recommendations"
    ]
)

4. Learning-Oriented Feedback

Principle : Focus on how the agent can improve, not blame

# ✅ GOOD: Learning-oriented
add_feedback(
    feedback_type="improvement",
    message="Analysis missed security vulnerability in auth flow",
    learning_opportunity="Add OWASP Top 10 checks to security analysis workflow",
    how_to_improve="Integrate security-auditor findings into code-analyzer reports"
)

# ❌ BAD: Blame-oriented
add_feedback(
    feedback_type="error",
    message="You failed to find the security issue",
    # No suggestions for improvement
)

Coordination Patterns

Pattern 1: Parallel Execution

When to Use : Multiple Group 1 agents can analyze simultaneously

# Orchestrator coordinates parallel Group 1 analysis
from lib.group_collaboration_system import coordinate_parallel_execution

results = coordinate_parallel_execution(
    group=1,
    agents=["code-analyzer", "security-auditor", "smart-recommender"],
    task_id=task_id,
    timeout_minutes=5
)

# All Group 1 findings consolidated before sending to Group 2
consolidated_findings = consolidate_findings(results)
send_to_group_2(consolidated_findings)

Pattern 2: Sequential Coordination

When to Use : Groups must execute in order (1→2→3→4)

# Standard workflow
findings = execute_group_1_analysis()  # Group 1: Analyze
plan = execute_group_2_decision(findings)  # Group 2: Decide
results = execute_group_3_execution(plan)  # Group 3: Execute
validation = execute_group_4_validation(results)  # Group 4: Validate

Pattern 3: Iterative Coordination

When to Use : Quality doesn't meet threshold, needs iteration

for iteration in range(max_iterations):
    # Group 3 executes
    results = execute_group_3(plan)

    # Group 4 validates
    validation = execute_group_4(results)

    if validation.quality_score >= 70:
        break  # Success!

    # Group 4 sends feedback to Group 2 for plan adjustment
    feedback = validation.get_improvement_suggestions()
    plan = group_2_adjust_plan(plan, feedback)

    # Group 3 re-executes with adjusted plan

Pattern 4: Conditional Coordination

When to Use : Execution path depends on analysis results

# Group 1 analysis
security_findings = security_auditor.analyze()

if security_findings.critical_count > 0:
    # Critical security issues → immediate path
    plan = group_2_create_security_fix_plan(security_findings)
    results = group_3_execute_security_fixes(plan)
else:
    # Normal path
    all_findings = consolidate_all_group_1_findings()
    plan = group_2_create_standard_plan(all_findings)
    results = group_3_execute_standard(plan)

Troubleshooting Collaboration Issues

Issue 1: Communication Not Reaching Target

Symptoms :

• Group 2 doesn't receive Group 1 recommendations
• Group 3 doesn't receive execution plan

Diagnosis :

from lib.group_collaboration_system import get_communications_for_agent

# Check if communications are recorded
comms = get_communications_for_agent("strategic-planner", communication_type="recommendation")
if not comms:
    print("❌ No communications found - sender may not be recording properly")

Fix :

• Ensure record_communication() is called after analysis
• Verify task_id is consistent across groups
• Check communication_type matches expected type

Issue 2: Feedback Loop Not Learning

Symptoms :

• Same mistakes repeated
• No improvement in recommendation confidence
• Agents don't adjust based on feedback

Diagnosis :

from lib.agent_feedback_system import get_feedback_stats

stats = get_feedback_stats("code-analyzer")
if stats["total_feedback"] == 0:
    print("❌ No feedback received - feedback loop broken")

Fix :

• Ensure Group 4 sends feedback after validation
• Verify agents query feedback before making decisions
• Check feedback is actionable and specific

Issue 3: Knowledge Not Transferring

Symptoms :

• Groups rediscover same patterns
• Best practices not reused
• Learning not retained

Diagnosis :

from lib.inter_group_knowledge_transfer import get_knowledge_transfer_stats

stats = get_knowledge_transfer_stats()
if stats["successful_transfers"] < stats["total_knowledge"] * 0.5:
    print("⚠️ Low knowledge transfer success rate")

Fix :

• Ensure agents query knowledge before tasks
• Add context matching to knowledge queries
• Increase knowledge confidence through successful applications

Issue 4: Group Specialization Not Developing

Symptoms :

• All agents perform similarly across task types
• No clear specialization patterns
• Sub-optimal task routing

Diagnosis :

from lib.group_specialization_learner import get_specialization_profile

profile = get_specialization_profile(group_num=3)
if not profile.get("specializations"):
    print("⚠️ No specializations detected - need more task diversity")

Fix :

• Record observations for all task executions
• Ensure task types are correctly labeled
• Allow sufficient tasks (50+) for specialization to emerge
• Review specialization insights regularly

Success Metrics

Effective Group Collaboration Indicators :

• ✅ Communication flow rate > 95% (messages reach intended recipients)
• ✅ Feedback loop cycle time < 5 minutes (validation → feedback → learning)
• ✅ Knowledge reuse rate > 60% (discovered patterns applied in future tasks)
• ✅ Recommendation effectiveness > 85% (Group 1 recommendations followed and successful)
• ✅ Execution success rate > 90% (Group 3 executes plans successfully first time)
• ✅ Validation pass rate > 80% (Group 4 validates without requiring major iterations)
• ✅ Specialization emergence rate: Each group develops 3+ specializations after 100 tasks

Track with:

from lib.group_collaboration_system import get_group_collaboration_stats

stats = get_group_collaboration_stats()
print(f"Communication success rate: {stats['communication_success_rate']:.1%}")
print(f"Average feedback cycle time: {stats['avg_feedback_cycle_seconds']}s")
print(f"Knowledge reuse rate: {stats['knowledge_reuse_rate']:.1%}")

Integration Examples

Example 1: Complete Four-Tier Workflow

# Orchestrator coordinates complete workflow
from lib.group_collaboration_system import record_communication
from lib.agent_feedback_system import add_feedback
from lib.inter_group_knowledge_transfer import query_knowledge, add_knowledge
from lib.group_specialization_learner import get_recommended_group_for_task

# Step 0: Get specialization recommendations
routing = get_recommended_group_for_task(
    task_type="refactoring",
    complexity="medium",
    domain="authentication"
)
print(f"Recommended: {routing['recommended_agents']}")

# Step 1: Group 1 analyzes (code-analyzer)
analysis = code_analyzer.analyze(task)

# Query existing knowledge
existing_patterns = query_knowledge(
    for_group=1,
    knowledge_type="pattern",
    task_context={"task_type": "refactoring", "domain": "authentication"}
)

# Send findings to Group 2
record_communication(
    from_agent="code-analyzer",
    to_agent="strategic-planner",
    task_id=task_id,
    communication_type="recommendation",
    data=analysis
)

# Step 2: Group 2 decides (strategic-planner)
user_prefs = preference_coordinator.load_preferences()
plan = strategic_planner.create_plan(analysis, user_prefs)

# Send plan to Group 3
record_communication(
    from_agent="strategic-planner",
    to_agent="quality-controller",
    task_id=task_id,
    communication_type="execution_plan",
    data=plan
)

# Step 3: Group 3 executes (quality-controller)
results = quality_controller.execute(plan)

# Send results to Group 4
record_communication(
    from_agent="quality-controller",
    to_agent="post-execution-validator",
    task_id=task_id,
    communication_type="execution_result",
    data=results
)

# Step 4: Group 4 validates (post-execution-validator)
validation = post_execution_validator.validate(results)

# Send feedback to Group 1
add_feedback(
    from_agent="post-execution-validator",
    to_agent="code-analyzer",
    task_id=task_id,
    feedback_type="success",
    message="Recommendations were 95% effective",
    details={"quality_improvement": 18}
)

# Send feedback to Group 3
add_feedback(
    from_agent="post-execution-validator",
    to_agent="quality-controller",
    task_id=task_id,
    feedback_type="success",
    message="Execution was efficient and effective"
)

# Share successful pattern
if validation.quality_score >= 90:
    add_knowledge(
        source_group=4,
        knowledge_type="pattern",
        title="Successful Authentication Refactoring Pattern",
        description=f"Pattern used in task {task_id} achieved quality score {validation.quality_score}",
        context={"task_type": "refactoring", "domain": "authentication"},
        evidence={"quality_score": validation.quality_score}
    )

References

Related Systems :

• lib/group_collaboration_system.py - Communication tracking
• lib/agent_feedback_system.py - Feedback management
• lib/inter_group_knowledge_transfer.py - Knowledge sharing
• lib/group_specialization_learner.py - Specialization tracking
• lib/agent_performance_tracker.py - Performance metrics

Related Documentation :

• docs/FOUR_TIER_ARCHITECTURE.md - Complete architecture design
• docs/FOUR_TIER_ENHANCEMENTS.md - Advanced features
• agents/orchestrator.md - Orchestrator coordination logic

Weekly Installs

0

Repository

bejranonda/llm-…r-claude

GitHub Stars

18

First Seen

Jan 1, 1970

Security Audits

Gen Agent Trust HubPassSocketPassSnykPass