sit with sid
Development Using Claude Code or Codex header artwork

Development Using Claude Code or Codex · Part 4

Part 4: Task Breakdown Skill

April 10, 2026 · 1 min read

Task BreakdownPlanningExecution

A good breakdown is what separates smooth execution from random coding.

Skill for This Post: task-breakdown

Create .ai/skills/task-breakdown/SKILL.md:

---
name: task-breakdown
description: Break approved PRD intake into atomic engineering tasks with dependencies, estimates, and done criteria.
---
 
# Task Breakdown
 
## Rules
- Tasks must be independently testable.
- No task should mix API, UI, and infra unless absolutely necessary.
- Every task needs explicit done criteria.
- Mark tasks as parallelizable or sequential.
 
## Required Output
For each task:
- id
- title
- scope
- dependencies
- estimate (S/M/L)
- risk (low/medium/high)
- done criteria
 
Also output:
1) critical path
2) parallel tracks
3) first task to start now

Prompt Pattern

Run task-breakdown on the approved PRD intake.
Keep tasks atomic and testable.
Output a numbered task list plus critical path.

Why This Skill Matters

  • Prevents hidden dependency surprises
  • Enables parallel work without merge chaos
  • Gives code review and QA clean handoff boundaries

Cross-Agent Install

ln -sfn "$PWD/.ai/skills/task-breakdown" ~/.claude/skills/task-breakdown
ln -sfn "$PWD/.ai/skills/task-breakdown" ~/.codex/skills/task-breakdown

Next

Part 5 uses the first task and executes it with a tight implementation loop.