task-scaler（任务缩放器）

Task Scaler

A technology-agnostic skill for evaluating task complexity and determining appropriate workflow scale.

Core Purpose

Classify tasks into appropriate scale categories to optimize:

• Agent utilization
• Workflow complexity
• Resource allocation
• Token consumption

Scale Classification

Categories

trivial:
  description: "Minimal changes with obvious implementation"
  characteristics:
    - Single-line or few-line changes
    - Typo fixes, whitespace corrections
    - Simple variable renames
    - Clear, obvious modifications
  metrics:
    lines_of_change: "< 3"
    files_affected: 1
    complexity_score: "< 5"
  workflow:
    direct_execution: true
    agents_required: 0
    rai_required: false
    deliverable_evaluation: false

small:
  description: "Single-component changes with clear scope"
  characteristics:
    - Single function implementation
    - Bug fix in one file
    - Simple feature addition
    - Minor refactoring
  metrics:
    lines_of_change: "3-50"
    files_affected: "1-3"
    complexity_score: "5-14"
  workflow:
    direct_execution: false
    agents_required: "1-2"
    rai_required: false
    deliverable_evaluation: true

medium:
  description: "Multi-component changes requiring coordination"
  characteristics:
    - Multiple function implementations
    - Cross-file changes
    - Feature with multiple components
    - Significant refactoring
  metrics:
    lines_of_change: "50-200"
    files_affected: "3-10"
    complexity_score: "15-29"
  workflow:
    direct_execution: false
    agents_required: "3-4"
    rai_required: true
    deliverable_evaluation: true

large:
  description: "Architectural changes with system-wide impact"
  characteristics:
    - Architecture modifications
    - New module or service
    - Multi-tenant considerations
    - System-wide impact
  metrics:
    lines_of_change: "200+"
    files_affected: "10+"
    complexity_score: "30+"
  workflow:
    direct_execution: false
    agents_required: "minimum needed"
    rai_required: true
    deliverable_evaluation: true
    full_workflow: true

Complexity Scoring

Scoring Factors

factors:
  file_count:
    weight: 2
    calculation: "2 points per affected file"

  dependency_depth:
    weight: 3
    calculation: "3 points per dependency level"

  test_requirement:
    weight: 5
    calculation: "5 points if tests needed"

  user_interaction:
    weight: 3
    calculation: "3 points if user input needed"

  integration_complexity:
    weight: 4
    calculation: "4 points per external integration"

  database_changes:
    weight: 5
    calculation: "5 points if schema changes"

Score Thresholds

thresholds:
  trivial: "score < 5"
  small: "5 <= score < 15"
  medium: "15 <= score < 30"
  large: "score >= 30"

Classification Algorithm

Step 1: Initial Classification

Parse user request for indicators:

trivial_indicators:
  keywords:
    - "typo", "fix typo", "correct spelling"
    - "whitespace", "formatting"
    - "single line", "one line"
  patterns:
    - Change target is explicit and simple
    - No logic changes required

small_indicators:
  keywords:
    - "add function", "implement method"
    - "fix bug", "resolve issue"
    - "single component", "one file"
  patterns:
    - Single component scope
    - Clear implementation path

medium_indicators:
  keywords:
    - "add feature", "implement"
    - "multiple components", "create tests"
    - "refactor"
  patterns:
    - Multiple files affected
    - Testing required

large_indicators:
  keywords:
    - "architecture", "system"
    - "new module", "new service"
    - "multi-tenant", "system-wide"
  patterns:
    - Architectural decisions needed
    - Broad impact scope

Step 2: Complexity Analysis

Analyze for complexity factors:
  1. Count estimated files affected
  2. Assess dependency depth
  3. Determine test requirements
  4. Identify user interaction needs
  5. Check for external integrations
  6. Evaluate database impact
  7. Calculate total score

Step 3: Context Adjustments

adjustments:
  scale_up_if:
    - "High integration with existing code"
    - "Ambiguous requirements"
    - "Multiple valid approaches"
    - "Security implications"

  scale_down_if:
    - "Established pattern/template exists"
    - "User provided detailed instructions"
    - "Similar change done recently"
    - "Well-documented requirements"

Output Format

scale_evaluation:
  task_scale: trivial|small|medium|large
  complexity_score: <number>

  factors:
    file_count: <number>
    dependency_depth: <number>
    test_required: true|false
    user_interaction: true|false
    integrations: <number>
    database_changes: true|false

  reasoning:
    initial_classification: "<based on keywords/patterns>"
    complexity_analysis: "<factor breakdown>"
    context_adjustments: "<any scale changes>"

  workflow_recommendation:
    agents_required: <number or range>
    rai_required: true|false
    parallel_possible: true|false
    estimated_iterations: <number>

Anti-Fragmentation Principles

Minimal Agent Usage

principle: "Use minimum agents needed for task"

by_scale:
  trivial: "Direct execution, no agents"
  small: "1-2 agents maximum"
  medium: "3-4 agents, batch similar work"
  large: "Minimum needed, maximize parallel"

anti_pattern: "Using 5+ agents for every task"

Purposeful Delegation

valid_delegation_reasons:
  - "Specialized skill required"
  - "Session isolation needed"
  - "Parallel processing benefit"

invalid_delegation_reasons:
  - "Just in case"
  - "For confirmation"
  - "Protocol says so"

Batch Similar Tasks

batching_principle: "Group similar tasks for single agent"

example:
  bad: "3 file fixes → 3 separate agents"
  good: "3 file fixes → 1 agent batch"

Integration

Used By

primary_users:
  - task-scale-evaluator: "Core skill for scale assessment"
  - "/dev-workflow command": "Workflow routing decisions"

Best Practices

1. Be Conservative: When uncertain, scale up
2. Consider Context: Same task varies by codebase familiarity
3. Avoid Over-Engineering: Match workflow to actual complexity
4. Review History: Similar tasks inform classification
5. Account for Risk: Security/data tasks scale up