Pepper Prompt Engineer for Codex

Turn a rough task into a usable prompt

Pepper Prompt Engineer 2.5.1 is a free, MIT-licensed skill for Codex. It builds or improves prompts using CRAFT+: role, task, context, success criteria, actions, constraints, reasoning mode, output format, examples and verification.

The workflow fills reasonable gaps through recorded assumptions, asks focused questions when necessary and formats the prompt for Claude, GPT, Gemini, DeepSeek or a universal target. It supports English and Russian. Prompts are in English by default with an explicit instruction for the response language.

Included

• The full prompt-construction workflow and model-format references.
• Conditional modules, clarification strategy and output-format rules.
• Worked prompt examples and a Python validator for structural checks.

Install and use

Extract the ZIP into ~/.agents/skills/, preserving its folder. Restart Codex and ask: “Use Pepper Prompt Engineer. Build a prompt for GPT to review these requirements and return ambiguities, acceptance criteria and missing information.”

This edition runs in Codex; the output prompt can target another model family. The workflow builds a prompt for another model or session to execute. Activate it explicitly when you want to write or improve a prompt.

Verification and limits

The package and SKILL.md passed validation. The bundled GPT example’s actual prompt body passed the packaged Python validator. These checks establish prompt structure, not correctness of a downstream model’s answer. No new end-to-end model evaluation was performed for this listing.

The Python validator is optional for basic use and recommended for complex/improved prompts. Python 3 is needed to run it. The 700-word prompt limit and bilingual policy are documented in the skill. Free download; your normal model usage costs still apply.