What problem does it solve?
GPT-5.3-Codex moves fast and tends to skip reading files, over-refactor, drift beyond scope, and add backwards-compatibility shims unless prompts are tightly constrained. This Skill provides ready-to-use prompt blocks that prevent those failure modes.
Core Features & Use Cases
- Modular prompt blocks: Copy-paste XML-tagged sections for output verbosity, scope constraints, context loading, plan-first mode, long-context handling, uncertainty management, user updates, and tool usage.
- Reasoning effort guidance: A table mapping task types (simple generation, refactors, code review) to the appropriate
model_reasoning_effort setting in Codex CLI.
- Behavioral steering tactics: Quick-reference tips like using "cutover" to prevent compatibility hedging, forcing file reads before questions, and steering mid-task.
- Use Case: You are writing a system prompt for a Codex 5.3 agent that refactors code across multiple files. Include the plan-first, scope constraints, and context loading blocks to keep it on task.
Quick Start
Ask the agent to draft a system prompt for GPT-5.3-Codex that enforces plan-first execution and strict scope constraints using the codex-5-3-prompting skill.
