---

name: controlled-task-decomposition description: Breaks specifications into executable, verifiable tasks without over-fragmentation. Use after spec freeze. triggers: [implementation-start, plan-generation, complex-task-received] outputs: [task-backlog, dependency-map]

Controlled Task Decomposition

Purpose

Converts a large "What" (Spec) into several manageable "Hows" (Tasks). This skill ensures that implementation is incremental, testable, and doesn't overwhelm the Agent's context.

When to use this skill

• During the transition from Planning to Execution
• When a task is found to be too complex to implement in one go
• When creating an implementation_plan.md

Decomposition Steps

1. Map Tasks to Spec Sections: Every task MUST implement at least one spec ID (e.g., TASK-1 implements AUTH-001).
2. Avoid Micro-Tasks: Don't create tasks for minor things like "add import". Combine them into meaningful functional units.
3. Define Acceptance Criteria: Every task needs a clear "When is this done?" statement.
4. Respect Dependencies: Order tasks so that foundation components are built before items that depend on them.

Decision Tree

Copyflowchart TD
    A[Start Decomposition] --> B{Task size > 200 lines?}
    B -->|Yes| C[Split into Functional Units]
    B -->|No| D{High Risk?}
    D -->|Yes| E[Isolate in Single Task]
    D -->|No| F{Independent?}
    F -->|Yes| G[Parallelize if possible]
    F -->|No| H[Sequencing Required]
    C --> B
    E --> I[Generate Task List]
    G --> I
    H --> I

Review Checklist

1. Traceability: Can I trace every task back to a spec requirement?
2. Verifiability: Is there a test that can prove this task is complete?
3. Context: Can the agent complete this task without needing 10 other open files?
4. Structure: Does it follow the task-template?

How to provide feedback

• Be specific: "Task #2 is 'Implement Auth', which covers 4 different spec IDs and 500 lines of code."
• Explain why: "Too much scope in one task leads to implementation errors and context loss."
• Suggest alternatives: "Split into 'Task 2a: JWT Logic' and 'Task 2b: Middleware Integration'."

Tasks must be independently verifiable.