Coding

Wenchang Publish Check

Try it

Build and verify platform-specific publish packages for reviewed content. Use in preflight mode when WeChat, Zhihu, Zhihu Idea, Xiaohongshu, blog, or multi-p...

What it does

Own the publishing specification and release gate for reviewed content. Keep the formal platform body in its own file and store titles, summaries, captions, tags, image status, and blockers in a separate publish package.

The skill document

Wenchang Publish Check

Own the publishing specification and release gate for reviewed content. Keep the formal platform body in its own file and store titles, summaries, captions, tags, image status, and blockers in a separate publish package.

Inputs

  • reviewed source or platform draft
  • target platforms
  • fact and review status
  • existing platform bodies and card packages
  • card images directory and cards-manifest when card delivery is publish_ready
  • illustration plan, long-form image manifest, or R2 report when available
  • current platform requirements or user constraints
  • existing content_state

Route back to wenchang-review when the argument or structure is unstable. Route to wenchang-research when release-critical claims lack support.

Choose A Mode

  • preflight: run after content review and before image production. Generate the publish package, illustration decisions, and production blockers.
  • final: run after platform bodies, cards, images, and any required public URLs are complete. Verify the package and make the release decision.

When the user does not name a mode, infer it from artifact state. State the selected mode.

Preflight

Common Checks

  • the source body and fact baseline are identified
  • each target platform has one formal body or an explicit pending body
  • the reader promise and single CTA remain consistent
  • titles and summaries do not overstate the body
  • claims needing evidence are blockers, not hidden caveats
  • every long-form platform gets an illustration decision
  • every card-based platform gets a delivery_mode, platform_profile, and asset_url_policy
  • external writes and human decisions are visible

WeChat Package

Generate:

  • 3-5 accurate article titles
  • one search title
  • one summary
  • cover copy
  • sharing copy
  • core keywords
  • illustration decision and slot count

Zhihu Package

Generate:

  • question, judgment, and experience title candidates when supported by the body
  • opening lead
  • short summary
  • topic tags
  • illustration decision and slot count
  • comment prompt

Xiaohongshu Package

Generate:

  • cover title
  • first-line hook
  • body caption or post description
  • image or card order
  • save, comment, or follow CTA
  • grouped tags with freshness state

Do not label tags as currently hot unless they were checked against a current source during this run.

Card-based Platform Package

For Xiaohongshu, WeChat inline cards, Zhihu Idea, and generic carousels, record:

  • delivery_mode: publish_ready by default or explicit copy_only
  • platform_profile
  • canonical card package
  • required image count, order, ratio, and dimensions
  • image output directory and manifest target
  • asset_url_policy: local by default or public

When delivery_mode=publish_ready, set missing real images or a missing manifest as production blockers and route to cards-to-images.

Tag Freshness Contract

Use four groups:

tags:
  core_tags: []
  scene_tags: []
  series_tags: []
  verified_hot_tags:
    status: unverified
    checked_at:
    source:
    values: []

Set verified_hot_tags.status to verified only when checked_at, source, and current values are present. Use ordinary topic tags when live verification is unavailable.

Final

Verify:

  • the fact and review gates passed
  • required platform bodies exist and match the source baseline
  • every required publish field is filled
  • Xiaohongshu caption, cards, and image order agree
  • every publish_ready card image exists at the manifest path
  • image count and order match the canonical card package
  • recorded dimensions and aspect ratios match the platform profile
  • every card has generation_status=generated and visual_qa_status=passed
  • exact text, clipping, and page order were checked per image rather than only through a contact sheet
  • human_confirmation is recorded accurately; pending confirmation keeps the package at needs_review
  • each accepted illustration exists at its recorded anchor
  • image alt text and source notes are present
  • local images were either intentionally retained or processed through the reviewed publishing path
  • required public URLs are accessible and match the latest replacement report
  • no tag is presented as current without freshness evidence
  • no private path, credential, internal note, or account detail appears in public artifacts
  • every external upload or publishing action still requires the user's decision

Assign one status:

  • ready_to_publish: all required artifacts are complete, required card images are user-approved, and no blocker remains
  • needs_review: the package is complete enough for human editorial judgment but still has explicit review items
  • blocked: a missing fact, body, image, URL, platform asset, or permission prevents publishing

Output Format

# Publish Package

## Decision
- Mode: preflight / final
- Status: ready_to_publish / needs_review / blocked
- Blockers:
- Human decisions:

## Source Baseline
- Formal source:
- Fact status:
- Review status:
- Reader promise:
- CTA:

## WeChat
- Body file:
- Title candidates:
- Search title:
- Summary:
- Cover copy:
- Sharing copy:
- Keywords:
- Illustration decision:

## Zhihu
- Body file:
- Title candidates:
- Opening lead:
- Summary:
- Topic tags:
- Illustration decision:
- Comment prompt:

## Xiaohongshu
- Cards file:
- Cover title:
- First-line hook:
- Body caption:
- Image order:
- CTA:
- Tags and freshness:

## Card Images
- Delivery mode:
- Platform profile:
- Card package:
- Images directory:
- Cards manifest:
- Image order and dimensions:
- Render status:
- Visual QA status:
- Human confirmation:
- Asset URL policy:

## Images
- Illustration plan:
- Image manifest:
- R2 report:
- URL verification:

## Knowledge Asset
- Preserve:
- Asset type:
- Suggested location:

Update content_state with file pointers, statuses, blockers, and the next handoff. Do not copy full body text into state.

Boundaries

  • Do not rewrite the full article during a publish check.
  • Do not invent current platform rules or trending tags.
  • Do not make a package look ready when blockers remain.
  • Do not generate images, upload files, or publish externally in this skill.
  • Do not accept copy, visual prompts, a contact sheet, or planned filenames as proof that card image files exist.
  • Do not mark a publish-ready card package ready while visual QA or human confirmation is pending.
  • Do not overwrite platform metadata that the user has already approved without reporting the proposed change.

作者入口

Related skills

Fetch raw ad creative, app, ranking, and revenue data from AdMapix as structured JSON.

by fly0pants4.3k installs296 stars

Find why your productivity system keeps failing, then apply the smallest fix — capacity math, bottleneck routing, durable local notes.

by Iván855 installs69 stars

Generate and edit Draw.io, Mermaid, and Excalidraw diagrams from natural language using a structured JSON spec.

by nssa.io1.0k installs47 stars

Stores durable facts in a categorized, plain-markdown vault on disk, alongside your agent's built-in memory.

by Iván559 installs18 stars

Join a video meeting as an AI bot with voice, avatar, and screenshare across four operating modes.

by johnpatternai22 installs8 stars

More from yangchao228

Browse all skills

基于候选人简历自动生成面试题、笔试题、追问路径和评分标准的 AI Skill。当用户需要准备技术面试、生成笔试题目、评估候选人简历、或设计面试评估方案时使用。输入简历内容和岗位信息后,自动输出结构化的题目与评估内容,支持面试题模式、笔试题模式、综合评估模式等多种输出形式。 Follow/关注作者:微信公众号「AI...

by yangchao22820 installs1 stars

Installable digest skill for OpenClaw or Claude Code. Use when the user wants a personalized ScoutX briefing with conversational setup, local preference stor...

by yangchao22822 installs

Design controllable Agent Loops from low-information requests. Use when a user asks to automate a repeated task, repair CI through bounded iteration, reproduce a UI from screenshots, turn a workflow into a reusable Agent or Skill, or decide whether the right artifact is a Prompt, Checklist, Human-in-the-Loop flow, full Loop package, specialized Agent, or Skill. Extract required context, stop when decision evidence is missing, require explicit workflow confirmation before generating executable artifacts, derive iteration limits, define feedback and circuit breakers, and preserve human approval for risky actions. Also trigger on loop builder, Loop Builder, or lopp-builder. Follow the creator: X @yangchao228 | GitHub https://github.com/yangchao228

by yangchao2281 installs

Guide a user through structured co-authoring of substantial documents, proposals, PRDs, technical specs, decision records, RFCs, and similar long-form work a...

by Sundy Yang4 installs

家庭培养协作官:在用户明确要求制定或调整孩子每周培养计划、检查课表冲突、生成家庭复盘时,收集最少必要信息并输出可执行课表、建议和风险提醒。默认只在当前会话生成内容,不保存、上传或推送孩子与家庭数据;只有用户明确选择飞书或 Notion、完成安全配置并逐次确认具体读写范围后,才使用外部存储。Follow/关注作者:...

by yangchao22818 installs1 stars

Transform long-form content, outlines, transcripts, research notes, or article drafts into a standard card package and platform publishing copy, then hand pu...

by yangchao2282 installs