sit with sid
Development Using Claude Code or Codex header artwork

Development Using Claude Code or Codex · Part 5

Part 5: Task Implementation Skill

April 11, 2026 · 1 min read

ImplementationTestingDelivery

Implementation quality improves dramatically when the agent is forced to stay on one scoped task.

Skill for This Post: task-implementation

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

---
name: task-implementation
description: Implement a single planned task with minimal scope drift, verification, and completion notes.
---
 
# Task Implementation
 
## Inputs
- task id
- task scope
- done criteria
- dependencies already completed
 
## Rules
- Work on ONE task only.
- Do not expand scope unless user approves.
- Write or update tests for changed behavior.
- Run relevant verification before declaring done.
 
## Required Output
1) files changed
2) behavior implemented
3) tests added/updated
4) verification results
5) known limitations

Prompt Pattern

Run task-implementation for TASK-003 only.
Do not start any other task.
Show file diff summary, tests, and verification output.

Why This Skill Matters

  • Reduces scope creep
  • Keeps PRs reviewable
  • Produces repeatable, auditable implementation notes

Cross-Agent Install

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

Next

Part 6 applies a dedicated code-review gate skill to catch defects before QA.