全流程微信公众号自动发文 skill

Automate a full local WeChat Official Account publishing workflow: environment setup, content drafting, image prep, publishing, archiving, and optional sched...

WeChat Auto Publishing Complete

Use this skill to reproduce, document, and operate a complete local WeChat Official Account auto-publishing workflow without embedding any secrets, private account details, or personal identifiers in the skill package.

This skill is intentionally broader than a minimal workflow note. It includes the operational context needed to reproduce the workflow on a fresh machine, while still keeping all sensitive values external.

Core outcome

The desired end state is a reusable local workflow that can do the following:

1. prepare the environment on a new machine
2. gather the day’s source material
3. determine the article angle
4. draft the article in a target style
5. prepare cover.png, image1.jpg, and image2.jpg
6. assemble a publishable Markdown package
7. publish to the WeChat draft box
8. optionally complete formal publication
9. archive outputs and execution results
10. optionally attach scheduling and alerting

Required safety rule

Never store real private values in this skill package.

Do not include:

• real WECHAT_APP_ID
• real WECHAT_APP_SECRET
• real GOOGLE_API_KEY
• any real cookies, tokens, session secrets, API keys, or private IDs
• user-specific公众号 identifiers unless the user explicitly asks for them to be hard-coded
• any private filesystem paths that reveal personal context unless rewritten as placeholders

When documenting configuration, only include:

• variable names
• placeholder values
• lookup order
• validation methods
• safe example file structures

Skill structure

Read the bundled references depending on what the user is trying to accomplish:

• references/environment-and-config.md — use when preparing a new machine or validating prerequisites
• references/source-gathering.md — use when collecting the day’s source pool and market angle
• references/writing-style.md — use when drafting and formatting the article
• references/image-strategy.md — use when preparing cover and body images, including gallery mode
• references/publishing.md — use when publishing to draft/final and archiving outputs
• references/scheduling-and-alerting.md — use when attaching cron, wrappers, logs, and alerts
• references/security-boundary.md — use when checking what the skill must and must not contain
• templates/article-template.md — use as the default publishable article skeleton
• templates/env.example.txt — use as the safe non-secret environment placeholder file
• templates/run.sh — use as a starting point for a local orchestrator script
• templates/run.production-example.sh — use as a more production-ready orchestrator template with gallery, publish mode, and time-slot support
• templates/cron.example.txt — use as a starting point for scheduling
• templates/publish-result.example.json — use as a result artifact template
• templates/gallery-config.example.txt — use as a local gallery config example
• templates/cover-image-extend.example.md — use as a safe cover-image preference example
• templates/image-gen-extend.example.md — use as a safe image-gen preference example
• templates/workspace-tree.txt — use as a recommended directory layout
• runbook.md — use as the operator-facing execution checklist

Standard execution flow

Step 1: Prepare environment

Before doing any content work, verify that the environment is reproducible on the target machine.

Confirm:

• runtime dependencies exist
• the publishing script dependency chain exists
• non-sensitive config placeholders are in place
• secrets will be supplied outside the skill package
• external prerequisites such as WeChat IP allowlisting are understood

Read references/environment-and-config.md.

Step 2: Gather source material

Collect raw source material, filter it down, and compress the market angle.

Read references/source-gathering.md.

Step 3: Draft the article

Draft an article in the target style using the publishable Markdown conventions from this skill.

Read references/writing-style.md and start from templates/article-template.md when useful.

Step 4: Prepare images

Prepare:

• cover.png
• image1.jpg
• image2.jpg

Choose image sources explicitly:

• user-provided assets
• local gallery mode
• generated images

Read references/image-strategy.md.

Step 5: Assemble the package

Make sure the article package is self-contained and publishable:

• article.md
• cover.png
• image1.jpg
• image2.jpg

Use relative paths and UTF-8 encoding.

Step 6: Publish to draft

Use the local publishing pathway to send the article to WeChat draft.

Do not treat publishing as successful unless the draft step returns a meaningful success result, such as a valid media_id.

If baoyu-post-to-wechat 因依赖问题（如 simple-xml-to-json 兼容性错误）无法运行，使用备用脚本：

bash

node templates/publish.mjs

备用脚本 publish.mjs 支持完整流程：获取 token -> 上传封面 -> 上传内图 -> Markdown 转 HTML -> 创建草稿 -> 正式发布 -> 归档结果。

Read references/publishing.md.

Step 7: Optionally complete formal publication

If the workflow includes final publication, submit, poll, and capture the final URL.

备用脚本 templates/publish.mjs 已内置正式发布功能（freepublish），如果使用备用脚本完成 Step 6，正式发布会自动执行，无需额外操作。

Step 8: Archive results

Save result artifacts, logs, identifiers, and remaining gallery state where relevant.

Step 9: Add scheduling and alerting

If the user wants timed operation, attach a wrapper script and scheduler entry.

Read references/scheduling-and-alerting.md.

Important real-world publishing boundary

freepublish is not always operationally equivalent to manual publishing in the MP backend

In practice, a workflow may successfully:

• submit draft publication
• obtain publish_id
• obtain article_id
• obtain article_url

but still not behave the same way as manually publishing the draft inside the WeChat Official Account admin console, especially for homepage visibility expectations.

Therefore, the skill must distinguish between:

• technical publication success
• platform backend success
• operational visibility success

Recommended production modes

Mode A: draft_only (recommended production default)

text

auto content generation
→ auto image preparation
→ auto draft submission
→ human publishes in MP backend

Use this mode when homepage visibility and platform-consistent display matter more than full automation.

Mode B: full_publish (testing / exploratory mode)

text

auto content generation
→ auto draft submission
→ auto freepublish submit
→ poll result
→ archive article_url

Use this mode only when the operator explicitly accepts that API publication may not be equivalent to backend manual publication.

Multi-account deployment principle

When operating multiple public accounts, prefer:

text

one account = one working directory = one .env = one title history = one cron entry

This avoids credential confusion, title-history pollution, log mixing, and accidental cross-account publishing.

Image validity principle

Do not trust file extension alone.

Before publish, validate that images are:

• present
• readable
• non-empty
• in a real supported format (PNG/JPEG/WebP if supported by the target chain)

If a file is named cover.png but is actually HEIF/HEVC, the WeChat API may reject it.

Writing quality principle

Do not directly inject scraped source titles into the body as-is.

Avoid:

• raw news-title insertion
• truncated title fragments ending with ...
• media-style prefixes such as 报道： in final body text

Prefer:

• title cleaning
• summary transformation
• author-tone rewriting

Workflow principles

Keep the package explicit

Prefer explicit files and paths over hidden assumptions.

Keep secrets external

Configuration structure belongs inside the skill. Real credentials do not.

Keep failure handling visible

If a step fails, preserve enough output and log context for a human operator to understand what happened.

Keep image behavior configurable

Do not hard-code a single image source assumption. The workflow should make it clear whether the body images come from a local gallery, generated assets, or user-supplied files.

Reproduction checklist

A fresh-machine reproduction should be able to answer yes to all of these:

• runtime tools are installed
• publishing dependency chain is installed
• safe config placeholders are present
• operator knows where to put real secrets outside the skill package
• source gathering pathway is defined
• article template is available
• image strategy is defined
• publishing checks are defined
• result archive format is defined
• scheduling examples exist if automation is needed

Final rule

This skill is allowed to be operationally complete, but it must remain secret-free. Every reusable process detail can go into the skill; every real credential and personal account detail must stay outside it.