xhs-publish-assistant

스타9

포크2

업데이트2026년 7월 16일 02:41

Prepare copy-ready Xiaohongshu/Rednote publish materials from a local XHS article package. Use when the user asks for "输出发布格式", "小红书发布格式", "准备发布", "二次编辑发布格式", or needs title/body/tags/output image directory/checks without browser automation or clicking publish.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

출처

Dirtytrii

Dirtytrii/codex-skills

GitHub 저장소 열기 Creator 저장소 보기

다운로드

XHS Publish Assistant

Purpose

Prepare the final copy-paste bundle for a Xiaohongshu note. This skill is a publishing formatter, not a publishing robot.

Use it after the XHS role has already produced or approved the title, body, tags, and card images. Do not use this skill to choose topics, rewrite the note, generate cards, scrape comments, or register post-publish metrics.

Hard Boundaries

Do not open or automate Xiaohongshu pages.
Do not click publish, save, update, or any final platform button.
If the user asks for browser automation, login checks, creator-center fill, automatic publishing blockers, content-data export, or final posting, route to $xhs-automation-publisher and keep the same explicit final-publish approval gate.
Do not generate a publish HTML panel.
Do not list every image path unless the user explicitly asks; output only the output\ directory.
Do not modify published article source files. For a re-edit, use or create a reedit-YYYY-MM-DD\ package and keep the original as history.
Do not invent missing title/body/tags. If extraction is uncertain, say exactly what is missing and ask the user or read the source file manually.

Workflow

Resolve the target article directory.
- Normal package: content/xhs/<date>/<article>/
- Re-edit package: content/xhs/<date>/<article>/reedit-YYYY-MM-DD/
- If the user only says "这篇", use the latest active XHS path from the conversation. If ambiguous, ask for the path.
Read the relevant source:
- Prefer reedit-note.md inside a re-edit directory.
- Otherwise read content-pack.md.
Before producing the final publish format, ensure title/body/caption copy has passed $humanizer-zh in the current role workflow. If it has not, load $humanizer-zh and polish only the public copy without changing facts, dates, prices, claims, account identity, or image order.
- If the note includes narrative prose, story fragments, or dialogue, use $story-deslop only for those narrative passages after facts are locked.
- Do not use either skill to invent missing copy. If source copy is incomplete, stop and report what is missing.
Run the helper when available:

$skillRoot = if ($env:CODEX_HOME) { Join-Path $env:CODEX_HOME "skills\xhs-publish-assistant" } else { Join-Path $HOME ".codex\skills\xhs-publish-assistant" }
node (Join-Path $skillRoot "scripts\build_publish_format.cjs") "<article-or-reedit-directory>"

Review the helper output against the source file. Fix obvious extraction mistakes manually before responding.
Return only the final publish format. Keep it easy to copy.

Required Output Format

Use this exact section order:

标题

<title>

正文

<body>

标签

<hashtags, one line, <= 10 tags>

配图目录

<absolute output directory ending with backslash>

发布前检查

账号：<account or 待确认>
图片数量：<N>
图片尺寸：<全部 1080x1440 | mixed details | 未检查>
标题长度：<N> / 20
标签数量：<N> / 10
文案去 AI 味：<已运行 humanizer-zh | 已运行 humanizer-zh + story-deslop | 待确认>
标题覆盖：<short judgment>
封面职责：<short judgment>
图片顺序：按 output 目录内 xhs-*.png 文件名顺序上传
状态：<可发布 | 需补齐>

For a re-edit, append one extra section:

二次编辑记录

这是已发布笔记二次编辑。
编辑前基线：<views/comments/likes/favorites/shares if known>
编辑后请告诉我实际保存时间，我会登记为 reedit_phase。

Checks

Tags must be 10 or fewer. If more than 10, trim or ask which to keep.
Publish title must be 20 characters or fewer. Prefer 12-18 characters when possible. Count each Chinese character, Latin letter, number, punctuation mark, and visible space as one character. If the title is over 20, shorten it or mark the package as 需补齐.
The image directory should be an absolute path to an output\ folder, not individual images.
Prefer xhs-*.png ordering for upload.
Check PNG dimensions when images exist. Expected card size is 1080x1440.
Account must be explicit when possible. In this repo, read content/xhs/accounts.md if the package does not state the account.
For published/re-edited notes, mention the phase split so later retros do not mix original and edited performance.

Relationship To Other XHS Skills

XHS role decides the content strategy, title direction, body, card copy, and publish recommendation.
humanizer-zh is required before this skill returns a final copy-ready publish bundle.
story-deslop is optional and only for narrative/story/dialogue passages, not normal Xiaohongshu marketing or analysis copy.
xhs-visual-director is the default upstream for new carousel visuals, cover redesigns, and full visual rewrites. It should produce the style judgment, visual master, confirmation image, and final card images before this skill checks the output directory.
Existing HTML/render workflows or guizang-social-card-skill may still be used for legacy packages or small local edits, but the final publish bundle should state whether the visual direction was confirmed.
cheat-on-content handles prediction, publish registration, retro, and rubric calibration after the user publishes.
xhs-comment-research reads and analyzes comments; this skill does not scrape comments.
xhs-automation-publisher handles authorized browser/CDP automation. It defaults to preview/fill and requires explicit user approval before final posting or interaction commands.

이 저장소의 다른 Skills

같은 저장소

agent-role-orchestrator

Dirtytrii/codex-skills

Route CEO-first multi-window work into owner-led role loops. Use for 总控/CEO、架构/CTO、内容主编、角色派发与复用、模型路由、来源窗口回调、role-windows 台账、skill 命中统计或 multi-window loop engineering。

2026-07-189

skill-system-governance

Dirtytrii/codex-skills

Audit and optimize a skill system with evidence. Use for skill architecture, token or context waste, role and prompt boundaries, routing hits or misfires, plugin packaging, registry/docs/source drift, duplicate workflows, or deciding whether a reusable change belongs in a skill. Runs deterministic checks first and applies only authorized, minimal fixes.

2026-07-189

cheat-on-content

Dirtytrii/codex-skills

用打分、盲预测、发布后复盘和 rubric 进化校准内容判断。适用于视频、文章、播客等可量化内容；触发词包括初始化、打分、预测、发布、复盘、推荐选题、抓热点、状态、找对标。首次使用先运行 cheat-init。

2026-07-169

cheat-predict

Dirtytrii/codex-skills

为最终稿创建不可变盲预测日志，并在已有预测时追加新版本。通过 cheat-score-blind 做隔离评分，再由主窗口复核落盘。触发词包括启动预测、start prediction、打分并预测、写预测日志。

2026-07-169

cheat-bump

Dirtytrii/codex-skills

提议并执行 rubric 或 bucket 升级。两种模式：**完整 rubric bump**（最高风险动作，5 步强制 + 跨模型审核）和 **--bucket-only 轻量重校**（只换 bucket 边界，不动 rubric 公式）。**Phase 2 强制走 cheat-score-blind sub-agent 给校准池重打分**——不接受 self-scored fallback。触发词："升级 rubric"/"bump rubric"/"更新公式"/"我想加一个维度"/"调整权重"/"重校桶"/"recalibrate bucket"。

2026-07-169

cheat-init

Dirtytrii/codex-skills

cheat-on-content 的首次 onboarding 与脚手架创建器。统一流程——所有用户都走相同 5 阶段闭环，唯一区别是"发过视频的人"会在 init 时多一步：抓取已有视频建立历史 context（用于后续 cheat-seed 给更贴合的选题、更准的 baseline）。触发词："初始化"/"init"/"首次使用"/"我是新用户"/"setup cheat-on-content"。**必须在用户第一次会话执行；其他子 skill 在 .cheat-state.json 不存在时自动路由到此。**

2026-07-169

name	xhs-publish-assistant
description	Prepare copy-ready Xiaohongshu/Rednote publish materials from a local XHS article package. Use when the user asks for "输出发布格式", "小红书发布格式", "准备发布", "二次编辑发布格式", or needs title/body/tags/output image directory/checks without browser automation or clicking publish.

XHS Publish Assistant

Purpose

Prepare the final copy-paste bundle for a Xiaohongshu note. This skill is a publishing formatter, not a publishing robot.

Hard Boundaries

Do not open or automate Xiaohongshu pages.
Do not click publish, save, update, or any final platform button.
If the user asks for browser automation, login checks, creator-center fill, automatic publishing blockers, content-data export, or final posting, route to $xhs-automation-publisher and keep the same explicit final-publish approval gate.
Do not generate a publish HTML panel.
Do not list every image path unless the user explicitly asks; output only the output\ directory.
Do not modify published article source files. For a re-edit, use or create a reedit-YYYY-MM-DD\ package and keep the original as history.
Do not invent missing title/body/tags. If extraction is uncertain, say exactly what is missing and ask the user or read the source file manually.

Workflow

Resolve the target article directory.
- Normal package: content/xhs/<date>/<article>/
- Re-edit package: content/xhs/<date>/<article>/reedit-YYYY-MM-DD/
- If the user only says "这篇", use the latest active XHS path from the conversation. If ambiguous, ask for the path.
Read the relevant source:
- Prefer reedit-note.md inside a re-edit directory.
- Otherwise read content-pack.md.
Before producing the final publish format, ensure title/body/caption copy has passed $humanizer-zh in the current role workflow. If it has not, load $humanizer-zh and polish only the public copy without changing facts, dates, prices, claims, account identity, or image order.
- If the note includes narrative prose, story fragments, or dialogue, use $story-deslop only for those narrative passages after facts are locked.
- Do not use either skill to invent missing copy. If source copy is incomplete, stop and report what is missing.
Run the helper when available:

$skillRoot = if ($env:CODEX_HOME) { Join-Path $env:CODEX_HOME "skills\xhs-publish-assistant" } else { Join-Path $HOME ".codex\skills\xhs-publish-assistant" }
node (Join-Path $skillRoot "scripts\build_publish_format.cjs") "<article-or-reedit-directory>"

Review the helper output against the source file. Fix obvious extraction mistakes manually before responding.
Return only the final publish format. Keep it easy to copy.

Required Output Format

Use this exact section order:

标题

<title>

正文

<body>

标签

<hashtags, one line, <= 10 tags>

配图目录

<absolute output directory ending with backslash>

发布前检查

账号：<account or 待确认>
图片数量：<N>
图片尺寸：<全部 1080x1440 | mixed details | 未检查>
标题长度：<N> / 20
标签数量：<N> / 10
文案去 AI 味：<已运行 humanizer-zh | 已运行 humanizer-zh + story-deslop | 待确认>
标题覆盖：<short judgment>
封面职责：<short judgment>
图片顺序：按 output 目录内 xhs-*.png 文件名顺序上传
状态：<可发布 | 需补齐>

For a re-edit, append one extra section:

二次编辑记录

这是已发布笔记二次编辑。
编辑前基线：<views/comments/likes/favorites/shares if known>
编辑后请告诉我实际保存时间，我会登记为 reedit_phase。

Checks

Tags must be 10 or fewer. If more than 10, trim or ask which to keep.
Publish title must be 20 characters or fewer. Prefer 12-18 characters when possible. Count each Chinese character, Latin letter, number, punctuation mark, and visible space as one character. If the title is over 20, shorten it or mark the package as 需补齐.
The image directory should be an absolute path to an output\ folder, not individual images.
Prefer xhs-*.png ordering for upload.
Check PNG dimensions when images exist. Expected card size is 1080x1440.
Account must be explicit when possible. In this repo, read content/xhs/accounts.md if the package does not state the account.
For published/re-edited notes, mention the phase split so later retros do not mix original and edited performance.

Relationship To Other XHS Skills

XHS role decides the content strategy, title direction, body, card copy, and publish recommendation.
humanizer-zh is required before this skill returns a final copy-ready publish bundle.
story-deslop is optional and only for narrative/story/dialogue passages, not normal Xiaohongshu marketing or analysis copy.
xhs-visual-director is the default upstream for new carousel visuals, cover redesigns, and full visual rewrites. It should produce the style judgment, visual master, confirmation image, and final card images before this skill checks the output directory.
Existing HTML/render workflows or guizang-social-card-skill may still be used for legacy packages or small local edits, but the final publish bundle should state whether the visual direction was confirmed.
cheat-on-content handles prediction, publish registration, retro, and rubric calibration after the user publishes.
xhs-comment-research reads and analyzes comments; this skill does not scrape comments.
xhs-automation-publisher handles authorized browser/CDP automation. It defaults to preview/fill and requires explicit user approval before final posting or interaction commands.