fearovex/output-style-creator

output-style-creator

Wizard that creates new output-styles following docs/output-style-guidelines.md. Asks the user 5 focused questions, then writes a compliant file under output-styles/.

Triggers: /output-style-create, create output style, new output style, generate output style, add output style

---

Process

Step 1 — Verify guidelines

Read docs/output-style-guidelines.md. If missing → stop and tell the user to create it first. This skill never operates without the spec.

Step 2 — Ask the name

One question:

What's the style name? (kebab-case, describes the experience — e.g. direct, concise, warm-direct. Do NOT use a character name.)

STOP and wait. Validate:

• kebab-case only.
• No existing file at output-styles/<name>.md. If it exists, ask whether to overwrite or pick a new name.
• Reject names that describe characters or roles (gentleman, pirate, sergeant, any human name). Suggest a tone-based alternative.

Step 3 — Ask the 5 wizard questions

Ask one question at a time. STOP after each and wait for the answer. Re-reading the prior answer before each question — never assume defaults silently.

Q1 — Language behavior

How should the style handle language?
  1. Adaptive: match the user's input language (recommended).
  2. Spanish only.
  3. English only.

Q2 — Register

Register in Spanish:
  1. Tuteo (default, recommended for technical work).
  2. Voseo (rioplatense; requires justification).
  3. Usted (formal; requires justification).

If user picks 2 or 3, ask a follow-up: "One-sentence reason this style needs voseo/usted instead of tuteo." Save the answer; it goes inline in the file.

Q3 — Pacing & density

Pacing and density:
  1. Concise: short sentences predominate, almost no paragraph longer than 3 lines.
  2. Balanced: short sentences with one longer one when needed for nuance.
  3. Expansive: full paragraphs allowed, exploratory.

Q4 — Disagreement posture

When the model disagrees with the user, how should it sound?
  1. Direct: state disagreement immediately, evidence, no softening.
  2. Direct with warmth (default): state disagreement, acknowledge user's reasoning,
     evidence.
  3. Diplomatic: lead with user's valid points, then disagree, then evidence.

Q5 — Closure

How should responses end?
  1. Contextual: short answers no close; debates one question; long answers
     one-sentence synthesis. (Recommended.)
  2. Fixed: same closing pattern every time. (Specify the pattern.)

If user picks 2, ask: "What's the fixed closing pattern? One short example."

Step 4 — Show preview, get confirmation

Render the proposed file content based on the answers. Use the skeleton in Step 5. Show it to the user inside a code block. Ask:

Confirm to write output-styles/<name>.md? (y / edit / n)

y → write. edit → ask which section to revise, loop back to that question. n → abort, leave nothing on disk.

Step 5 — Write the file

Use this exact skeleton. Fill placeholders from the answers. Keep total under 60 lines (hard limit per guidelines Principle 4).

---
name: <name>
description: <one-line experience description — not a character>
keep-coding-instructions: true
---

# <Name> Output Style

> <One-line experience description, same as frontmatter description.>

## Language

<Adaptive / Spanish only / English only — and the trigger rule.>

## Register

<Tuteo / Voseo / Usted in Spanish. Contractions yes/no in English.>
<If voseo or usted: one-line justification.>

## Pacing & density

<Concise / Balanced / Expansive. State sentence-length preference and paragraph
limit. State that no decorative sections are added.>

## Emphasis

<Bold for key terms. Italics for soft emphasis. Inline code for identifiers. CAPS
only for true intensity, never decorative.>

## Disagreement & closure

Disagreement: <Direct / Direct with warmth / Diplomatic — one sentence on how it
sounds in practice.>

Closure: <Contextual / Fixed — one sentence describing the rule. If fixed, give one
short example.>

Step 6 — Post-write reminders

After writing the file, print to the user:

Created: output-styles/<name>.md (<N> lines).

Next steps:
  1. Review the diff.
  2. To activate this style, set "outputStyle": "<name>" in settings.json.
  3. Run bash install.sh to deploy.

Do NOT modify settings.json. Do NOT run install.sh. Both are user decisions.

Step 7 — Save to engram

Call mem_save with:

• title: Created output-style <name>
• type: config
• topic_key: output-styles/<name>
• content: structured What/Why/Where/Learned noting the 5 wizard answers.

---

Rules

• One question at a time. STOP after each. Never bundle questions.
• Never invent answers. If the user is silent, wait — do not default.
• Never include personality, biography, philosophy, behavior rules, or technical opinions in the generated file. Reject those if the user offers them — point to CLAUDE.md instead.
• Never exceed 60 lines in the generated file. If the answers would produce a longer file, trim during preview and tell the user what was cut.
• Never write the file before Step 4 confirmation.
• Never modify settings.json or run install.sh — that's user-only.
• The five mandatory sections (Language, Register, Pacing & density, Emphasis, Disagreement & closure) MUST all appear. Skipping one is a hard failure.