I Listed a WeChat Publishing Skill on ClawHub: Why md2wechat 2.0 Feels More Complete

I recently listed md2wechat 2.0 on ClawHub, the official OpenClaw skill directory.

On the surface, that sounds like a normal release update.
In practice, it marks a bigger shift:

md2wechat is no longer just a Markdown-to-WeChat formatting utility. It is becoming a more complete content workflow for AI Agents.

This article focuses on three questions:

• what changed in md2wechat 2.0
• why the ClawHub listing matters
• how to get from installation to your first WeChat draft in Claude Code

Screenshot: md2wechat 2.0 and ClawHub

What actually changed in 2.0

The short version is this:

md2wechat 2.0 moves closer to a real Agent-facing publishing workflow instead of stopping at formatting.

Many tools can solve one isolated step. They may format text, upload something, or generate one image. But the full chain still depends on scattered scripts, manual glue, and a lot of guessing.

What I wanted in 2.0 was not just "more features."
I wanted the whole thing to feel more usable.

That meant tightening:

• the install path
• capability discovery
• image generation
• command clarity for agents
• the path from content generation to draft publishing

Screenshot: What changed in 2.0

The image workflow finally feels like part of the product

One of the clearest upgrades in this release is the image layer.

A lot of WeChat tools can handle text conversion, but the workflow breaks as soon as you need:

• a cover image
• an infographic
• a summary visual
• a style-controlled illustration

That is where automation often falls apart.

So 2.0 puts more weight on image generation as a first-class step. It now supports flows around:

• article covers
• infographics
• banner-style summary graphics
• hand-drawn looks
• Apple Keynote-style visuals
• Victorian engraving-style visuals

The goal is not just more styles. The goal is to make images part of a stable publishing chain.

In other words:

formatting, images, uploads, and draft creation now fit together much more naturally.

Screenshot: Image workflow and presets

Installation and documentation now describe the same reality

Another important part of 2.0 is that the install story is much less confusing.

Users usually do not fail on one command. They fail on uncertainty:

• what should I install first
• is the skill the same thing as the CLI
• does Claude Code use the same path as OpenClaw
• what should Obsidian and Claudian users do

So a large part of this work was not glamorous. It was documentation and path cleanup.

The main route is now much clearer:

• Mac users install the CLI
• Claude Code users install the CLI first, then the skill
• OpenClaw users follow a separate platform path
• Obsidian + Claudian has its own guidance

The point is not to write longer docs.
The point is to let a new user know which path actually applies to them.

Agents can discover capabilities before they act

One belief I keep coming back to is simple:

if a tool is meant for agents, it should not rely on guessing.

If an agent has to infer supported themes, image providers, prompts, and modes by scraping the README, the product surface is still too weak.

That is why 2.0 puts more structure around discovery.

The workflow can now start by inspecting:

• capabilities
• providers
• themes
• prompts

That changes the behavior of the system in a practical way:

agents can check what exists first, then decide how to run the task.

That is less flashy than a new command, but much more useful in real workflows.

Why the ClawHub listing matters

It is easy to think of a directory listing as just another distribution channel.

For md2wechat, it matters more than that.

ClawHub is part of the path from "repo utility" to "installed skill inside the OpenClaw ecosystem."

That changes several things at once:

• users do not have to start from the GitHub repo
• the install entry is easier to find
• platform metadata becomes clearer
• the workflow fits more naturally into OpenClaw usage patterns

That changes the first impression of the product:

is this just a repository with scripts, or is it a skill people can actually install and use?

Screenshot: md2wechat on ClawHub

The review boundary forced an important design decision

The listing process was not frictionless.

The key issue was that my initial design let the skill download the CLI during installation.

From a convenience perspective, that sounds fine.
From a marketplace security perspective, it is the wrong boundary.

The reason is straightforward:

a skill should not quietly fetch third-party executables during install.

That turns convenience into risk.

So I changed the structure:

• the CLI is installed separately
• the Skill handles knowledge, guidance, and workflow orchestration

That split made the product cleaner and also made the listing path more acceptable.

If you are building something similar, this is the practical lesson:

separate the skill from the runtime early, instead of treating them as one blob.

Screenshot: ClawHub review note

Screenshot: Split skill and runtime

If you want to try it today, start with Claude Code

md2wechat 2.0 now spans multiple environments, but I still think Claude Code is the best first stop for most users.

The reasons are practical:

• the install path is shorter
• the feedback loop is faster
• failures are easier to debug

A simple first-run flow looks like this.

Step 1: install the CLI

brew install geekjourneyx/tap/md2wechat

Screenshot: Install md2wechat CLI

Step 2: install the skill

npx skills add https://github.com/geekjourneyx/md2wechat-skill --skill md2wechat

One detail matters here:

this installs the skill, not the CLI.

So the order is important: install the CLI first, then the skill.

Screenshot: Install md2wechat skill 1

Screenshot: Install md2wechat skill 2

Screenshot: Install md2wechat skill 3

Step 3: initialize the config

md2wechat config init

You will at least want to configure:

1. the WeChat Official Account AppID and AppSecret
2. an image provider

If you want the cover flow to work on day one, it helps to connect an image service early. ModelScope is a reasonable beginner-friendly option.

Screenshot: Init config

Screenshot: WeChat developer config

Screenshot: ModelScope API key

Step 4: ask the agent to create and publish the first draft

In Claude Code, you can give it a prompt like:

From there the flow can continue through:

• writing
• formatting
• image generation
• asset upload
• draft creation

Once you see that run through end to end, the difference between earlier versions and 2.0 becomes much more concrete.

Screenshot: Claude Code writing flow

Screenshot: Draft flow 1

Screenshot: Draft flow 2

Screenshot: Draft created 1

Screenshot: Draft created 2

How this article differs from the existing 2.0 release post

There is already a more structural 2.0 release post on the site.
This article intentionally covers a different angle.

That earlier post is more about:

• why 2.0 is a real consolidation point
• what changed across the CLI, prompts, images, installers, and docs

This one is more about:

• why listing on ClawHub matters
• what practical constraint showed up during review
• where a new user should start now

Together they make a better series than one article trying to do everything.

What comes next

The ClawHub listing is only one step in the larger direction.

I still want to keep breaking this workflow down by real usage environment:

• how to connect writing, formatting, images, and drafts inside Claude Code
• how to run a WeChat workflow with one instruction inside OpenClaw
• how to connect local writing in Obsidian + Claudian to WeChat publishing

The point is not one impressive command.
The point is whether an AI Agent can actually carry the content workflow from start to finish.

Closing

The journey from "it can format" to "it participates in a full Agent workflow" is the direction I care about.

This year, the direction is very clear for me:

build tools for agents.

Not only interfaces for humans to click, but reusable capabilities that agents can call inside real workflows.

That is why getting md2wechat 2.0 onto ClawHub feels like a meaningful milestone.

Public links:

• GitHub: geekjourneyx/md2wechat-skill
• Website: md2wechat.com
• ClawHub: clawhub.ai/geekjourneyx/md2wechat

Original WeChat article: I listed my WeChat publishing skill on ClawHub