---

name: task-management description: Task decomposition, dependency analysis, and parallel execution orchestration. Use when breaking down features into tasks, analyzing task dependencies, planning parallel execution strategies, or coordinating multi-agent workflows. Covers dependency detection, execution grouping, resource conflict prevention, and performance optimization.

Task Management Skill

Purpose

Provide systematic approaches for decomposing complex features into executable tasks, analyzing dependencies, and orchestrating parallel execution for optimal performance.

Key Capabilities:

• Feature decomposition strategies
• Dependency graph analysis
• Parallel execution planning
• Resource conflict detection
• Performance optimization

---

When to Use This Skill

Auto-activates when:

• Planning feature implementation (breaking down PRPs)
• Analyzing task dependencies (what depends on what)
• Orchestrating multi-agent execution (parallel vs sequential)
• Optimizing workflow performance (speedup calculations)
• Detecting resource conflicts (file access, shared state)

---

Quick Start

Feature Decomposition Checklist

• Identify deliverables: What files/features must exist?
• Extract tasks: One task per cohesive unit (1 file or tightly related files)
• Detect dependencies: Which tasks must complete before others?
• Group by parallelism: Which tasks can run simultaneously?
• Assign domains: Which expert handles each task?
• Validate completeness: Do tasks cover entire feature?

Dependency Analysis Checklist

• File-level: Which files create vs. consume?
• Data-level: Which outputs feed into inputs?
• Tool-level: Which commands must run sequentially?
• Domain-level: Which expertise areas must coordinate?

Parallel Execution Checklist

• Independence check: Tasks don't modify same files
• Grouping: Max 2-6 tasks per parallel group
• Context preparation: All contexts ready BEFORE invocation
• Single response: All Task() calls in ONE message
• Performance validation: Timing confirms parallelization

---

Core Principles (5 Key Rules)

1. Single Responsibility Per Task

Each task should accomplish ONE cohesive goal:

✅ GOOD - Focused tasks
Task 1: Create user model (src/models/user.py)
Task 2: Create user API endpoints (src/api/users.py)
Task 3: Create user tests (tests/test_users.py)

❌ BAD - Monolithic task
Task 1: Implement entire user system (10 files, mixed concerns)

Why: Parallel execution, clear validation, easier debugging

2. Explicit Dependency Tracking

Dependencies must be explicit, not assumed:

✅ GOOD - Clear dependencies
Task 2: Create user API
DEPENDS_ON: [Task 1]  # Needs user model
FILES_CONSUMED: src/models/user.py

❌ BAD - Implicit dependencies
Task 2: Create user API
# Silently expects user model to exist

Why: Enables dependency-aware grouping, prevents failures

3. Parallel Groups Must Be Independent

Tasks in same group cannot conflict:

✅ GOOD - Independent tasks
Group 1 (parallel):
  - Task 1: Create src/models/user.py
  - Task 2: Create src/models/post.py
  - Task 3: Create src/models/comment.py
# Different files, no conflicts

❌ BAD - Conflicting tasks
Group 1 (parallel):
  - Task 1: Create src/api/routes.py (lines 1-50)
  - Task 2: Modify src/api/routes.py (lines 51-100)
# Same file, race condition!

Why: File conflicts, race conditions, corrupted output

4. Context Preparation Before Invocation

Prepare ALL contexts BEFORE making any Task() calls:

✅ GOOD - Parallel execution pattern
# 1. Prepare contexts
ctx1 = f"Task 1: Create {file1}..."
ctx2 = f"Task 2: Create {file2}..."
ctx3 = f"Task 3: Create {file3}..."

# 2. Invoke all in SAME response
Task(subagent_type="expert1", prompt=ctx1)
Task(subagent_type="expert2", prompt=ctx2)
Task(subagent_type="expert3", prompt=ctx3)
# Time = max(T1, T2, T3)

❌ BAD - Sequential invocation
for task in tasks:  # Loop = sequential!
    Task(subagent_type="expert", prompt=task)
# Time = T1 + T2 + T3 (3x slower)

Why: Parallel execution (3x speedup), efficiency

5. Validate Performance Gains

Measure actual speedup from parallelization:

✅ GOOD - Performance validation
sequential_time = sum([5, 4, 5])  # 14 minutes
parallel_time = max([5, 4, 5])    # 5 minutes
speedup = (14 - 5) / 14 * 100     # 64% faster

if speedup < 50:
    print("WARNING: Expected 50%+ speedup, got {speedup}%")

Why: Confirms parallelization working, detects sequential execution bugs

---

Common Patterns

Pattern 1: Linear Dependency Chain

Use Case: Each task builds on previous (A → B → C → D)

Strategy: Sequential execution groups

Group 1: Task A (foundation)
Group 2: Task B (depends on A)
Group 3: Task C (depends on B)
Group 4: Task D (depends on C)

# No parallelization possible
# Total time = TA + TB + TC + TD

Pattern 2: Independent Parallel Tasks

Use Case: No dependencies between tasks (A, B, C independent)

Strategy: Single parallel group

Group 1 (parallel):
  - Task A (src/models/user.py)
  - Task B (src/models/post.py)
  - Task C (src/models/comment.py)

# Time = max(TA, TB, TC) ≈ 33% of sequential

Pattern 3: Parallel with Shared Dependency

Use Case: Multiple tasks depend on same foundation (A → B, A → C, A → D)

Strategy: Sequential foundation, parallel dependents

Group 1: Task A (foundation - e.g., database schema)
Group 2 (parallel):  # All depend on A completing
  - Task B (user API)
  - Task C (post API)
  - Task D (comment API)

# Time = TA + max(TB, TC, TD)

Pattern 4: Multi-Level Dependency Tree

Use Case: Complex dependency graph

Strategy: Topological sort + grouping

Group 1: [A, B]          # No dependencies, run parallel
Group 2: [C]             # Depends on A
Group 3: [D, E]          # D depends on B, E depends on C
Group 4: [F]             # Depends on D and E

# Time = max(TA,TB) + TC + max(TD,TE) + TF

---

Quick Reference

Dependency Types

Type	Example	Detection
File Creation	Task 2 imports Task 1's file	Read file list
Data Flow	Task 2 uses Task 1's output	Read task descriptions
Tool Order	terraform apply after terraform plan	Domain knowledge
Domain Order	Backend before frontend (API first)	Architecture convention

Parallelization Limits

Scenario	Max Parallel Tasks	Reason
Optimal	2-6 tasks	Claude API limits
File conflicts	1 task per file	Race conditions
Shared state	1 task	Data corruption
Complex coordination	3 tasks	Diminishing returns

Performance Calculation

# Sequential time
sequential = sum(task_durations)

# Parallel time
parallel = max(task_durations)

# Speedup percentage
speedup = (sequential - parallel) / sequential * 100

# Efficiency (ideal = 100% per core)
efficiency = speedup / num_parallel_tasks * 100

---

Anti-Patterns to Avoid

❌ Anti-Pattern 1: Over-Granular Tasks

Problem: 50 micro-tasks for simple feature Issue: Coordination overhead > execution time Fix: Combine related changes into cohesive tasks

❌ Anti-Pattern 2: Mega-Tasks

Problem: One task creates 20 files Issue: No parallelization, unclear validation Fix: Decompose into single-responsibility tasks

❌ Anti-Pattern 3: Hidden Dependencies

Problem: Task silently expects file from previous task Issue: Race conditions, failures in parallel execution Fix: Explicit DEPENDS_ON declarations

❌ Anti-Pattern 4: Parallelizing Dependent Tasks

Problem: Running Task B before Task A completes (B depends on A) Issue: Task B fails, wasted execution time Fix: Dependency-aware grouping (sequential groups)

❌ Anti-Pattern 5: Sequential Invocation in Loop

Problem: for task in tasks: Task(...) Issue: 3x slower than parallel (defeats purpose) Fix: All Task() calls in single response

---

Workflow Example

Scenario: Implement User Authentication Feature

Step 1: Decompose Feature

Deliverables:
  - User model with password hashing
  - Authentication API endpoints
  - JWT token management
  - Login/logout/refresh routes
  - Integration tests

Tasks:
  Task 1: Create User model (src/models/user.py)
  Task 2: Create auth service (src/services/auth.py)
  Task 3: Create JWT utilities (src/utils/jwt.py)
  Task 4: Create auth endpoints (src/api/auth.py)
  Task 5: Create tests (tests/test_auth.py)

Step 2: Analyze Dependencies

Task 1 (User model):
  DEPENDS_ON: []
  CREATES: src/models/user.py

Task 2 (Auth service):
  DEPENDS_ON: [Task 1]  # Imports User model
  CREATES: src/services/auth.py

Task 3 (JWT utils):
  DEPENDS_ON: []  # Independent utility
  CREATES: src/utils/jwt.py

Task 4 (Auth endpoints):
  DEPENDS_ON: [Task 2, Task 3]  # Uses auth service + JWT
  CREATES: src/api/auth.py

Task 5 (Tests):
  DEPENDS_ON: [Task 4]  # Tests complete system
  CREATES: tests/test_auth.py

Step 3: Group by Parallelism

Group 1 (parallel):
  - Task 1 (User model)
  - Task 3 (JWT utils)
# Independent tasks, different files

Group 2 (sequential):
  - Task 2 (Auth service - depends on Group 1)

Group 3 (sequential):
  - Task 4 (Auth endpoints - depends on Group 2)

Group 4 (sequential):
  - Task 5 (Tests - depends on Group 3)

# Time = max(T1,T3) + T2 + T4 + T5
# vs Sequential: T1 + T2 + T3 + T4 + T5
# Speedup: ~20% (limited by dependency chain)

Step 4: Execute with Prepared Contexts

# Group 1 - Parallel execution
ctx1 = f"Create User model at src/models/user.py..."
ctx3 = f"Create JWT utilities at src/utils/jwt.py..."

Task(subagent_type="python-backend-expert", prompt=ctx1)
Task(subagent_type="python-backend-expert", prompt=ctx3)

# Groups 2-4 run sequentially after Group 1 completes

---

Navigation Guide

Need to...	Read this
Analyze complex dependencies	dependency-analysis.md
Optimize parallel execution	parallel-execution.md

---

Resource Files

dependency-analysis.md

Comprehensive dependency detection algorithms, graph analysis, conflict resolution strategies

parallel-execution.md

Advanced parallelization patterns, performance benchmarking, optimization techniques

---

Skill Status: COMPLETE ✅ Line Count: 412 ✅ Progressive Disclosure: 2 resource files ✅