draft-xhs-caption

draft-xhs-caption

Take a transcript (or topic + bullet notes) and produce a 小红书 caption draft for the user to review. Stop at the draft; do not publish.

When this skill applies

• User explicitly asks for a draft / 草稿 / 起草 / 写文案.
• User pastes a cleaned transcript and wants a caption from it.
• User says "帮我想个小红书标题" or similar.

When this skill does NOT apply

Hand off to xiaohongshu or post-to-xhs instead when the user says:

• "发小红书", "发到小红书", "发布", "上传"
• "post to xhs", "publish to RedNote"
• Anything that implies the act of posting, not writing.

If the user's request is ambiguous, ask: "你是想我先起草一版文案给你看，还是直接发？"

Output schema

Always emit a single fenced markdown block in this exact shape:

## 标题
<≤20 个汉字，第一人称「我」语气>

## 正文
<200–400 字。短句、空行多。第一人称「我」。不要用「我们 AI 从业者…」之类的圈层语气。>

## 话题
#标签1 #标签2 #标签3 #标签4 #标签5

Voice rules

Lifted from AGENTS.md → 小红书 caption voice guidelines:

• Use 「我」, not 「我们」.
• Avoid in-group / elitist framing (例如「我们 AI 从业者」「懂的都懂」).
• Share an observation, not a lesson; no preachy tone.
• 4–6 hashtags. Mix one broad (#播客, #小红书日常) with topic-specific ones.

Process

1. Confirm with the user: "用这段当输入，对吗？要 draft 几版？默认 1 版。"
2. If they want N versions, emit N fenced blocks, each labelled ### Draft 1 etc.
3. After emitting, ask: "选一版？还是要我再改某一处？"
4. Stop. Do not call the posting skill. If the user then says "发", direct them to the xiaohongshu / post-to-xhs skill.