planning-and-task-breakdown

Planning and Task Breakdown

Overview

Turn ambiguous or large goals into a sequence of small, independently verifiable tasks. Good planning dramatically increases the success rate of agent-driven work.

When to Use

• Starting any non-trivial feature or change
• User asks for a plan, breakdown, tasks, or implementation steps
• Before beginning coding on a new area
• When a previous attempt failed due to scope or unclear requirements

Core Process

1. Clarify the Goal

Restate the objective in one or two sentences. Surface assumptions explicitly:

ASSUMPTIONS:
1. ...
2. ...
→ Correct me if any of these are wrong.

2. Identify Scope Boundaries

What is explicitly in scope? What is out of scope for this iteration?

3. Decompose into Vertical Slices

Prefer thin vertical slices (end-to-end value) over horizontal layers when possible.

Each task should be:

• Small enough to complete and verify in one focused session
• Independently testable
• Ordered by dependency

4. Write Clear Acceptance Criteria

For every task, define how we will know it is done (tests, behavior, observable outcome).

5. Sequence and Estimate Roughly

Order tasks. Note any that can be parallelized.

Output Format

Produce a structured plan:

## Goal
[One sentence]

## Assumptions
- ...

## Out of Scope
- ...

## Tasks

### Task 1: [Short title]
**Description**: ...
**Acceptance criteria**:
- [ ] ...
- [ ] ...
**Dependencies**: none / Task X

### Task 2: ...

Quality Checks

• Are tasks small enough?
• Does each have verifiable acceptance criteria?
• Is the first task a good starting point (low risk, high learning)?
• Have major risks or unknowns been called out?

Common Pitfalls

• Tasks that are still too large ("Implement the whole auth system")
• Missing acceptance criteria
• Hidden assumptions not surfaced
• Starting implementation before the plan is agreed

Verification

A good plan lets another agent (or human) pick up any task and know exactly what "done" looks like without further clarification.