Workflow Patterns

A technology-agnostic skill for managing multi-agent workflows via team orchestration.

Core Patterns

Team Orchestration Pattern

pattern: team_orchestration description: "Create team, define tasks with dependencies, spawn workers" use_when:

• "Non-trivial tasks (small/medium/large)"
• "Multiple agents needed"
• "Quality gates required"

implementation: 1_create_team: "TeamCreate with descriptive name" 2_create_tasks: "TaskCreate with blockedBy dependencies" 3_spawn_workers: "Task tool with team_name to spawn teammates" 4_monitor: "Watch TaskList and teammate messages" 5_evaluate: "Task(deliverable-evaluator) when chain complete" 6_cleanup: "SendMessage(shutdown_request) + TeamDelete"

Task Dependency Pattern

pattern: task_dependencies description: "Declare execution order via blockedBy relationships" use_when:

• "Steps must execute in order"
• "Some steps can run in parallel"

example: tasks: - id: 1 subject: "Implement feature" blockedBy: [] - id: 2 subject: "Run tests" blockedBy: [1] - id: 3 subject: "Review code" blockedBy: [1] - id: 4 subject: "Evaluate deliverables" blockedBy: [2, 3]

Tasks 2 and 3 run in parallel after task 1

Task 4 waits for both 2 and 3

Iteration Pattern

pattern: iteration description: "Retry via new tasks until success or max iterations" use_when:

• "Quality gates must pass"
• "Debugging cycles needed"
• "Refinement required"

implementation: on_test_failure: 1: "Create debug task (test-failure-debugger)" 2: "Create fix task (code-developer, blockedBy: debug)" 3: "Create retest task (test-executor, blockedBy: fix)"

on_evaluation_fail: 1: "Create fix tasks from DE recommendations" 2: "Create retest + re-review tasks" 3: "Create re-evaluate task (blockedBy: retest, re-review)"

exit_conditions: - "DE returns PASS" - "Max iterations (3) reached" - "User intervention requested"

Direct Delegation Pattern

pattern: direct_delegation description: "Simple Task call without team overhead" use_when:

• "Trivial tasks"
• "Single agent needed"
• "No coordination required"

implementation: single_task: "Task(code-developer) with full context" no_team: true no_task_list: true

Workflow Templates

Standard Implementation Workflow

template: standard_implementation scale: small|medium

phases: analysis: - Task(goal-clarifier) → acceptance_criteria - Task(task-scale-evaluator) → scale, subtasks

execution: - TeamCreate("dev-workflow") - TaskCreate: implement (no deps) - TaskCreate: test (blockedBy: implement) - TaskCreate: review (blockedBy: implement) - TaskCreate: evaluate (blockedBy: test, review) - Spawn: code-developer, test-executor, quality-reviewer

evaluation: - Task(deliverable-evaluator) when tasks complete - On FAIL: create fix tasks, loop - On PASS: cleanup team

cleanup: - SendMessage(shutdown_request) to all teammates - TeamDelete

Full Feature Workflow

template: full_feature scale: large

phases: analysis: - Task(goal-clarifier) → acceptance_criteria - Task(task-scale-evaluator) → scale, subtasks, parallel_groups

design: - Task(design-architect) → architecture spec

execution: - TeamCreate("dev-workflow") - TaskCreate per subtask with dependency graph - Spawn: code-developer, test-developer, test-executor, quality-reviewer

evaluation: - Task(deliverable-evaluator) - Retry loop (max 3)

cleanup: - Shutdown teammates, TeamDelete

Debug Loop

template: debug_loop trigger: "Test failures during execution"

steps: 1_analyze: - TaskCreate: "Debug test failures" - Assign to test-failure-debugger (or use Task delegation)

2_fix: - TaskCreate: "Apply fix" (blockedBy: debug task) - Assign to code-developer

3_verify: - TaskCreate: "Re-run tests" (blockedBy: fix task) - Assign to test-executor

4_check: - If pass: continue to review/evaluate - If fail: loop (max 3 iterations)

Coordination Strategies

Progress Tracking via TaskList

tracking: use: "TaskList tool to check status" states: pending: "Not yet started" in_progress: "Being worked on" completed: "Done"

monitoring: - "Check TaskList after teammate messages" - "Identify blocked tasks" - "Assign unblocked tasks to idle teammates"

Inter-Agent Communication

communication: teammate_updates: via: "SendMessage(type: message)" when: "Task completion, blocking issues, questions"

team_announcements: via: "SendMessage(type: broadcast)" when: "Critical issues only (expensive)"

shutdown: via: "SendMessage(type: shutdown_request)" when: "All work complete, DE returns PASS"

Error Handling

error_handling: test_failure: action: "Create debug + fix + retest tasks" max_retries: 3

agent_failure: action: "Re-spawn teammate, reassign task"

blocked_task: detection: "Task stuck in pending with unresolved blockedBy" action: "Investigate blocking task, escalate if needed"

evaluation_fail: action: "Create targeted fix tasks from DE feedback" max_iterations: 3

Anti-Patterns

Over-Orchestration

anti_pattern: over_orchestration description: "Using team for trivial tasks" symptom: "TeamCreate for single-line change" solution: "Use direct Task delegation for trivial work"

Missing Exit Conditions

anti_pattern: infinite_loop description: "Retry without max iteration check" symptom: "Endless fix-test-fail cycle" solution: "Always enforce max 3 iterations, escalate to user"

Premature Evaluation

anti_pattern: premature_evaluation description: "Calling DE before implementation chain completes" symptom: "DE fails because tests haven't run" solution: "Use blockedBy to ensure evaluate waits for test + review"

Integration

Used By

primary_users:

• "/dev-workflow command": "Orchestration patterns"

secondary_users:

• task-scale-evaluator: "Workflow recommendations"