promptify

It's pretty sure you're running Sonnet 5. It'll ask anyway.

---

A skill that drafts polished prompts — or Claude Code /goal completion conditions, where applicable — tuned to whichever Claude or OpenAI model family is currently running. Neither Anthropic nor OpenAI gives a skill any reliable way to read the active model ID at runtime, and their own docs say so. promptify doesn't pretend otherwise: it guesses, tells you what it guessed, and takes a correction if you give one.

Before / after

You paste a rough idea into whatever session you happen to have open. The agent drafts something generic — maybe it reaches for a chain-of-thought instruction that actively hurts an o-series model, maybe it invents a temperature parameter that 400s on Sonnet 5, maybe it fabricates a /goal line for Codex CLI, which has no such mechanism.

With promptify:

Detected: Claude Code, Claude Sonnet 5 — say so if wrong.

...then the actual drafting happens against that specific model family's rules — not a generic average of every model's prompting advice.

Install in Claude Code

/plugin marketplace add albert-mr/promptify

Registers this repo as a plugin marketplace so Claude Code can find and install promptify.

/plugin install promptify@promptify

Installs the promptify plugin from that marketplace into your Claude Code environment.

/promptify:promptify

Invokes the skill to turn your rough idea into a polished prompt or /goal completion condition.

Install in Codex CLI

promptify currently ships to Codex as a direct agent skill. Codex scans .agents/skills from your current working directory up to the repo root, and it also scans $HOME/.agents/skills. Symlinked skill folders are supported.

• This repo: no install step is needed. The checked-in .agents/skills/promptify symlink points at skills/promptify, so Codex picks it up when you start in this repo or one of its subfolders.

• Another project: from this checkout, symlink it into that project's skill directory:

  mkdir -p /path/to/project/.agents/skills
  ln -s "$(pwd)/skills/promptify" /path/to/project/.agents/skills/promptify

• Global install: from this checkout, symlink it into your user skill directory:

  mkdir -p ~/.agents/skills
  ln -s "$(pwd)/skills/promptify" ~/.agents/skills/promptify

Use cp -R skills/promptify ... instead of ln -s ... if you want a fixed copy rather than live updates from this checkout. Invoke the skill in Codex with $promptify; if it does not appear immediately, restart Codex.

Repo layout

• skills/promptify/SKILL.md — the skill definition itself.
• skills/promptify/references/*.md — supporting reference docs:
  • claude-families.md
  • openai-families.md
  • goal-vs-normal.md
  • detection-fallbacks.md
• .claude-plugin/ — Claude Code packaging metadata (marketplace + plugin manifest).
• .agents/skills/promptify — symlink used for Codex CLI's repo-level skill discovery.
• .github/workflows/validate.yml — CI check that validates the plugin manifests and skill files on every PR.

Scope and honesty notes

• No provider (Anthropic or OpenAI) gives reliable runtime model-ID introspection — model-family detection here is best-effort and its guess is always disclosed and correctable, never asserted as fact.
• Codex CLI has no equivalent of Claude Code's /goal completion-condition primitive, so goal-mode prompt drafting is Claude-Code-only; on Codex, promptify only produces normal (system/task/agent) prompts.
• When in doubt about which model family is actually running, trust your own knowledge of your environment over promptify's guess and correct it.

See MAINTAINING.md for the versioning policy, the update/PR process (including the monthly automated freshness check), and the regression checklist any change must pass.

License

MIT © Albert Martinez