Codex Prompt Optimizer

Overview

Use this skill to turn a human request into a high-signal prompt for OpenAI Codex or Codex CLI. The goal is not to make prompts longer. The goal is to make them easier for Codex to execute correctly: outcome-first, scoped, grounded in real repo facts, and verifiable.

Source guidance distilled from OpenAI Prompt guidance for GPT-5.5 and Codex prompting docs:

• Prefer shorter, outcome-first prompts over process-heavy prompt stacks.
• Define what good looks like: target outcome, success criteria, constraints, available evidence, final response shape.
• Avoid unnecessary always / never / must rules unless they are true invariants.
• Give Codex enough context and exact validation commands when possible.
• Break complex work into focused tasks; smaller changes are easier to test and review.
• Add explicit stopping conditions and missing-evidence behavior.
• Ask Codex to verify its work with the cheapest relevant check.

When to Use

Use when: