Majkey25

Ask about my work

Answers use my public projects, profile and approved public sources.

Questions and attachments go to OpenAI through Cloudflare. OpenAI may use shared content to train models. Attachments are not saved in local history. Do not share sensitive data. Privacy

~ / work / 04

Prompt engineering skill

Turns tasks into prompts and agent rules with built-in verification criteria, for Claude Code, Codex and ChatGPT

area
AI and language
stack
Python · Markdown
as of
Sep 2026

Problem

Most prompt collections gather clever phrases. A prompt a team relies on needs a clear contract and a way to check what comes back.

Approach

The skill turns a rough request into a prompt contract (objective, context, task, constraints, output format, verification and failure handling), following current OpenAI, Anthropic and Google guidance. Coding prompts are routed by the repository evidence that is available instead of invented details.

Result

Installs as a skill in ChatGPT, Codex and Claude Code. It ships a heuristic prompt linter and a generator for eval starter manifests, and for important prompts it pushes eval cases and iteration instead of guessing.