---

model: opus created: 2025-12-16 modified: 2025-12-16 reviewed: 2025-12-16 name: agent-coordination-patterns description: | Coordinate multi-agent workflows: sequential, parallel, and iterative patterns. Defines agent handoffs, dependencies, communication protocols, and integration. Use when designing multi-agent workflows, coordinating agent handoffs, planning agent dependencies, or building complex agent pipelines.

Agent Coordination Patterns

Description

Coordination patterns for sequential, parallel, and iterative agent workflows. Defines how agents work together, communicate findings, and maintain context across handoffs in multi-agent systems.

When to Use

Automatically apply this skill when:

• Designing multi-agent workflows
• Coordinating agent handoffs
• Planning agent dependencies
• Integrating agent outputs
• Building complex workflows
• Optimizing agent collaboration

Core Principle: Hybrid Context Sharing

Two complementary layers work together:

1. Transparency Layer (File-Based)

• Human-inspectable files
• Real-time progress visibility
• Easy debugging and inspection
• Clear agent coordination

2. Intelligence Layer (Knowledge Graph)

• Historical learning
• Pattern recognition
• Cross-session persistence
• Audit trails

Agent Integration Protocol

Phase 1: Pre-Execution Context Reading

Before starting work, agents MUST:

1. Read workflow context → Understand overall objective
2. Check agent queue → Know dependencies and position
3. Review shared data → Get requirements and standards
4. Read dependency outputs → Build on previous work

Example Flow:

CopyUser delegates to python-developer:

1. Read current-workflow.md → "Building REST API"
2. Read agent-queue.md → "research-assistant completed"
3. Read inter-agent-context.json → "Tech: FastAPI + PostgreSQL"
4. Read research-assistant-output.md → "Requirements defined"
5. Begin implementation with full context

Phase 2: Progress Reporting During Execution

Continuously communicate state:

• Update progress file every 5-10 minutes
• Report current activity clearly
• List completed steps
• Note any blockers immediately
• Estimate remaining time

Phase 3: Post-Execution Output Writing

Produce standardized results:

• Document all accomplishments
• Record technical decisions
• List created artifacts
• Note known issues
• Provide handoff guidance

Coordination Patterns

Sequential Coordination

Agents work one after another, each building on previous work.

Pattern Structure:

CopyAgent A completes → Writes output
  ↓
Agent B reads A's output → Starts work → Writes output
  ↓
Agent C reads A's and B's outputs → Starts work

When to use:

• Clear dependencies exist
• Each step requires previous completion
• Linear workflow progression

Example Workflow:

CopyResearch Assistant (30 min)
  ↓ (passes requirements)
Python Developer (90 min)
  ↓ (passes implementation)
Test Architect (45 min)
  ↓ (passes test suite)
Documentation Writer (30 min)

Best Practices:

• Complete handoff notes are critical
• Each agent validates previous work
• Include "next agent needs" section
• Document all assumptions made

Parallel Coordination

Multiple agents work simultaneously on independent tasks.

Pattern Structure:

Copy         ┌→ Agent A → Output A
Start →  ├→ Agent B → Output B  → Integration Agent
         └→ Agent C → Output C

When to use:

• Independent work streams
• No direct dependencies
• Can merge results later
• Want faster completion

Example Workflow:

CopyRequirements Defined
    ├→ Backend Developer (API)
    ├→ Frontend Developer (UI)
    └→ Database Architect (Schema)
         ↓
    Integration Tester (Verify all parts work together)

Best Practices:

• Define clear interfaces upfront
• Use shared contracts (API specs)
• Regular sync points
• Integration agent validates compatibility

Shared Contract Example:

Copy{
  "api_contract": {
    "/api/users": {
      "GET": "returns user list",
      "POST": "creates user"
    }
  },
  "data_models": {
    "User": {
      "id": "string",
      "name": "string",
      "email": "string"
    }
  }
}

Iterative Coordination

Agent revisits work based on feedback from other agents.

Pattern Structure:

CopyAgent A → Agent B (reviews) → Issues found
  ↑                               ↓
  ←───────── Feedback ────────────↓

When to use:

• Quality improvement cycles
• Refinement needed
• Review and feedback loops
• Progressive enhancement

Example Workflow:

CopyDeveloper → Code Review → Issues Found
    ↑                         ↓
    ←── Fix Issues ──────────↓

Security Auditor → Vulnerabilities Found
    ↑                         ↓
Developer ←── Apply Fixes ───↓

Best Practices:

• Clear feedback format
• Specific actionable items
• Track iteration count
• Define completion criteria

Feedback Format:

Copy## Review Feedback

### Critical Issues (Must Fix)
1. SQL injection vulnerability in `/api/users` line 45
2. Missing authentication on DELETE endpoint

### Improvements (Should Fix)
1. Add input validation for email format
2. Implement rate limiting

### Suggestions (Could Improve)
1. Consider caching for performance
2. Add more descriptive error messages

Hybrid Coordination

Combines multiple patterns for complex workflows.

Example Structure:

CopySequential Start:
  Research → Architecture Design
           ↓
Parallel Development:
  ├→ Backend Team
  ├→ Frontend Team
  └→ QA Test Planning
           ↓
Sequential Integration:
  Integration → Testing
           ↓
Iterative Refinement:
  Review ↔ Fixes

When to use:

• Complex projects
• Multiple teams
• Different phases need different patterns
• Optimization opportunities

Coordination Rules

Dependency Management

Hard Dependencies (Must Complete First):

Copy{
  "agent": "test-architect",
  "depends_on": ["python-developer"],
  "reason": "Cannot test code that doesn't exist"
}

Soft Dependencies (Preferred Order):

Copy{
  "agent": "documentation-writer",
  "prefers_after": ["test-architect"],
  "reason": "Better docs with test examples"
}

Communication Protocols

Status Broadcasting:

• STARTING: Beginning work
• IN_PROGRESS: Active work (% complete)
• BLOCKED: Cannot continue
• COMPLETED: Finished successfully
• FAILED: Could not complete

Handoff Requirements:

• Summary of work done
• Decisions made
• Artifacts created
• Known issues
• Next steps guidance

Conflict Resolution

When agents disagree:

1. Document both perspectives
2. Escalate to user if critical
3. Use precedence rules
4. Security > Functionality > Performance

Precedence Rules:

CopySecurity Auditor > All Others (security issues)
Architect > Developers (design decisions)
Senior > Junior (experience hierarchy)
Later > Earlier (recent context)

Best Practices

1. Clear Communication

• Explicit handoff notes
• Document all assumptions
• State dependencies clearly
• Update progress frequently

2. Robust Error Handling

• Check for previous failures
• Validate inputs exist
• Handle missing dependencies
• Report blockers immediately

3. Maintain Context

• Read all relevant outputs
• Preserve decision history
• Update shared context
• Avoid duplicating work

4. Quality Gates

• Validate before handoff
• Test integration points
• Review critical paths
• Ensure completeness

Common Pitfalls

• ❌ Starting without reading context
• ❌ Not updating progress during execution
• ❌ Vague or incomplete handoff notes
• ❌ Missing dependency outputs
• ❌ Parallel work without contracts
• ❌ No integration validation
• ❌ Infinite iteration loops

Integration with Other Skills

• agent-file-coordination: Uses file structures for coordination
• multi-agent-workflows: Higher-level workflow orchestration

References

• Related Skills: agent-file-coordination, multi-agent-workflows
• Design Patterns: Sequential, Parallel, Iterative, Hybrid
• Replaces: agent-context-management (coordination sections)