copywriter-curto

name	copywriter-curto
description	Generates short-form copy under tight character limits including hooks, captions, and ad headlines
version	0.1.0

Copywriter Curto

Short-form copy specialist. Produces hooks (under 10 words), captions (under 220 characters), and ad headlines (under 40 characters) tuned to the brand voice. Always routes through llm-router with task: caption so the cheapest qualified provider runs.

When to invoke

• Any task that asks for a hook, caption, headline, button label, push notification, or one-line tagline.
• First draft of social posts before they are expanded by caption-multi-platform.
• Ad creative iteration where headline is the variant axis.
• Email subject lines and preview text under 60 characters.
• Card titles, banner overlays, and any UI surface with a hard character limit.

Inputs

• kind: string. hook, caption, ad_headline, subject_line, cta_button, tagline.
• context: string. What the copy is about (product, theme, angle).
• audience: string. Target persona summary.
• tone: string. Brand tone descriptor (bold, intimate, expert, playful).
• language: string. pt-BR, en, es.
• max_chars: integer, optional. Override the default limit for kind.
• variants: integer. Number of options to return (default 5).
• must_include: array of strings, optional. Words or phrases that must appear.
• must_avoid: array of strings, optional. Banned words.

Process

1. Resolve the character limit: use max_chars if set, else the default for kind.
2. Compose a prompt that pins the kind, context, audience, tone, language, limit, must_include, must_avoid, and the requested number of variants.
3. Call llm-router with task: caption, sending the composed prompt.
4. Parse the response into a list of variants candidates.
5. Filter out any candidate that exceeds max_chars, fails must_include, or hits must_avoid.
6. If fewer than variants candidates pass, ask the LLM for more and merge results.
7. Score each surviving candidate on hook strength, clarity, and brand fit using a rubric.
8. Return the candidates sorted by score, plus the rubric scores.

Outputs

• candidates: array of objects. Each { text, chars, score, rubric: { hook, clarity, brand_fit } }.
• provider_used: string. Echoed from llm-router.
• cost_estimate_usd: number.

Examples

Example 1: hook for a reel about color analysis

Input: { kind: "hook", context: "color analysis service launch", audience: "brazilian women 28-45", tone: "expert confident", language: "pt-BR", variants: 5 } Output: { candidates: [{ text: "Sua cor te trai. Descubra qual e.", chars: 32, score: 0.91, rubric: {...} }, ...] }

Example 2: ad headline under 40 characters

Input: { kind: "ad_headline", context: "online wardrobe consult", audience: "young professionals", tone: "bold", language: "en", max_chars: 40, variants: 6 } Output: 6 headlines, all under 40 characters, sorted by score.

Failure modes

• All candidates exceed the limit: re-prompt with a stricter limit baked into the system message and a worked example.
• must_include word never appears: try one more pass with the term capitalized and bracketed in the prompt; if still absent, surface a warning.
• LLM returns fewer than requested: top up with another call and merge.
• Provider quota error: llm-router falls back automatically; surface which provider actually ran.

Related skills

• caption-multi-platform: takes the winning candidate from this skill and fans it out per platform.
• revisao-humanizada: humanizes the chosen candidate before publish.
• llm-router: provider routing and logging.
• compliance-generic: confirms the chosen copy passes vertical-agnostic compliance.
• compliance-<active-client>: optional vertical-specific audit when the host project ships its own compliance skill.