Task Decomposition

Use when complexity is

L

or

XL

. Goal: reduce risk before implementation starts.

---

Steps

1. Identify Boundaries

• List all modules and files involved.
• Map dependencies between them.
• Identify shared state or data contracts.

2. Split into Sub-Problems

• Each sub-problem is a self-contained unit of work.
• Give each one its own scope, approach, and done condition.
• Minimize cross-dependencies.

3. Determine Parallelism

• Independent sub-problems can run in parallel.
• Dependent sub-problems need explicit order.
• Shared state or shared files require coordination.

4. Create Sub-Plans

Each sub-problem gets a mini plan:

• Goal (one sentence).
• Files to modify.
• Validation approach.
• Exit criteria.

5. Assign

• Each sub-problem can be delegated to a separate agent.
• Provide full context: goal, files, constraints, and validation.
• Parent agent merges results and runs final validation.

---

Output Format

## Sub-Problem 1: [title]

Goal: ...
Files: ...
Depends on: none / Sub-Problem N
Parallel: yes / no
Validation: ...
Exit criteria: ...

## Sub-Problem 2: [title]

Goal: ...
Files: ...
Depends on: none / Sub-Problem N
Parallel: yes / no
Validation: ...
Exit criteria: ...

---

Rules

• Each sub-problem must be completable independently.
• Shared state changes must be coordinated.
• Sub-problems that modify the same file cannot be parallel.
• Use phases instead of parallelism when risk is high but independence is weak.