# Better Prompt Text

> Use when the user invokes /better-prompt-text or wants a rough prompt (Spanish or English) turned into a polished, precise English prompt to copy and use elsewhere — text delivery only, no execution, no files.

- Skill: `presidenteog/better-prompt-text` (Agent Skill)
- Install (CLI): `npx skillmds@latest add presidenteog/better-prompt-text`
- Raw SKILL.md: https://api.skillmd.com/api/skills/presidenteog/better-prompt-text/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: PresidenteOG (https://skillmd.com/u/presidenteog)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/presidenteog/better-prompt-text

---


# /better-prompt-text

Rough prompt → interrogation → excellent English prompt → deliver the text. Nothing else: no files, no execution, no council. Ends when the text is delivered.

## Usage

```
/better-prompt-text <rough prompt>     # input Spanish or English
```

## Gate

**Questions before rewrite — always.** Hard stop; overrides any keep-going bias (e.g. Auto Mode). Silent inference = violation.

## Interrogation

**Coverage:** 8 categories: goal · scope · constraints · tech stack · inputs/outputs · edge cases · success criteria · out-of-scope.

- Resolved = **unambiguous** from prompt or known context/memory; inapplicable → `n/a`; else → question. Yield ~10 simple, 12–15 complex; trivial (<~30 min) → 2–3 riskiest only, rest `[assumed]`. Never zero.
- **Delivery: AskUserQuestion dialogs only**, batches of ≤4, in the user's input language. Open questions get 2–4 plausible options (Recommended first); "Other" catches free text. Never chat lists.
- **Answers:** "don't know" → sensible default + `[assumed]`. Contradicts the rough prompt → state conflict + winner explicitly (answer wins). "enough, rewrite now" (any point) → `[assumed]` the rest → rewrite.

## Output

Always English, fenced block the user can copy:

```
## Objective            — one paragraph: what "done" looks like
## Context              — project, environment, prior work
## Requirements         — numbered, testable
## Constraints          — tech, conventions, limits
## Acceptance criteria  — checklist
## Out of scope         — explicit exclusions
```

Defaults shown `[assumed: …]`. After delivering: offer **edit / more questions** once; apply edits directly and re-present. Done. (User-initiated edit requests after Done are still honored — just not re-offered.)

## Red flags

- "Prompt looks complete, skip questions" → confirm the 2–3 riskiest anyway.
- "Guess silently" → every guess marked `[assumed]`.
- "n/a because hard" → n/a = inapplicable, not inconvenient.
- Output drifting into Spanish → English only.
- Any urge to persist a file, run the prompt, or dispatch agents → wrong tool; that's /better-prompt.

## Boundaries

Text out, nothing else. Works in any workspace (no dependencies). A managed copy also exists on claude.ai (anthropic-skills bundle); in desktop sessions where both are visible, this user-level copy is canonical — behavior is identical, never invoke both. Editing this file requires re-uploading the managed copy to claude.ai to keep them in sync.