LS WeChat Article Writer

Write and publish WeChat Official Account articles without manual formatting. This skill keeps workflow rules in documentation and repeatable execution in code. Use SKILL.md for trigger rules, execution modes, routing overview, resilience rules, and non-negotiable quality bars. Use references/ files for detailed execution guidance.

Workflow judgment stays in the docs and the agent. Toolkit commands should execute explicit decisions, not infer article strategy from content.

Onboarding

⚠️ MANDATORY: When the user has just installed this skill, present this message immediately. Do not ask whether they want an introduction. Translate it to the user's language.

✅ AI WeChat Article Writer installed!

Tell me your topic, or paste a Markdown draft, and I will turn it into a WeChat article.

Try it now:

• Help me write a WeChat article about AI coding
• Format this Markdown for WeChat and publish it to drafts
• Preview this WeChat article with the latepost-depth theme

What it does:

• Plans article angles from a topic or trending signals（Supports TrendRadar MCP）
• Writes professional articles with a de-AI voice
• Formats content with WeChat-optimized themes
• Generates optional cover and inline images
• Publishes directly to the WeChat draft box

One-time setup:

1. cd toolkit && npm install && npm run build && cd ..
2. pip install -r requirements.txt
3. mkdir -p .ls-wechat-article && cp config.example.yaml .ls-wechat-article/config.yaml
4. Fill in wechat.appid and wechat.secret, then add your public IP to the WeChat API whitelist

No WeChat API yet? You can still write and preview locally.

Usage

This skill supports four common entry patterns:

• Start from a concrete topic and draft a full article
• Start from existing Markdown and handle formatting, preview, or publishing only
• Start from a specific workflow step with --step
• Run post-publish maintenance such as analytics backfill, edit learning, or playbook refresh

Examples:

• Write a WeChat article about AI coding trends
• Format this Markdown for WeChat and publish it to drafts
• Start from --step 3.5 and help me choose a framework
• Show me the last 7 days of article performance

Setup

Before first use, prepare the environment:

1. Install Node dependencies and build the toolkit:

   cd toolkit && npm install && npm run build && cd ..
   

2. Install Python dependencies:

   pip install -r requirements.txt
   

3. Create the runtime config:

   mkdir -p .ls-wechat-article
   cp config.example.yaml .ls-wechat-article/config.yaml
   

4. Fill in these fields in .ls-wechat-article/config.yaml or ~/.liusir-skills/ls-wechat-article/config.yaml:

   • wechat.appid
   • wechat.secret

5. Add the current public IP to the WeChat API whitelist:

   curl -s https://ifconfig.me
   

Without WeChat API credentials, the skill can still draft, format, and preview locally, but it cannot publish to the draft box.

Optional: TrendRadar

If TrendRadar is installed, the skill can use a configured TrendRadar MCP client to fetch topic signals. If it is not configured, the skill should fall back to general hotspot scripts or direct user-provided topics.

TrendRadar is optional and is not required for drafting, formatting, preview, or publishing.

Skill Directory

Read files on demand. Do not load everything upfront.

Path	Purpose
references/pipeline.md	End-to-end execution detail
references/operations.md	First-run setup, client onboarding, publishing support, analytics, and learning workflows
references/writing-guide.md	Writing quality bar and de-AI rules
references/article-archetypes.md	Archetype routing and shape binding
references/editorial-qa.md	Step 5 report template and blocking rules
references/rewrite-examples.md	Few-shot rewrite patterns for Step 5A
references/frameworks.md	Article framework selection
references/topic-selection.md	Topic scoring and angle selection
references/seo-rules.md	Title, digest, tags, and SEO polish
references/visual-prompts.md	Cover and inline image prompt guidance
references/theme-selection.md	Public theme recommendation matrix
references/style-template.md	Client style template
references/wechat-constraints.md	WeChat rendering constraints
references/cli-reference.md	CLI command reference
references/skill-maintenance.md	Skill maintenance and validation rules
{runtime_root}/clients/{client}/style.yaml	Client voice, default theme, and image defaults
{runtime_root}/clients/{client}/history.yaml	Publish history and analytics
{runtime_root}/clients/{client}/playbook.md	Client-specific writing playbook
toolkit/dist/*.js	Built CLI entrypoints
scripts/*.py	Topic and SEO helper scripts

Execution Modes

Auto

• Move the workflow forward without pausing by default
• Stop only when required input is missing, a step and its fallback both fail, or the workflow reaches an explicit user-decision boundary
• If the user explicitly says 仅排版 / 仅发布 / 不要改内容, skip directly to Step 7
• If the user provides raw Markdown without that restriction, skip drafting but still run Step 5A and Step 5B before Step 7
• If the user provides a concrete topic, skip hotspot mining and topic generation
• Before image generation, ask once about image scope, style, and inline image density unless the user already specified them
• Before preview or publish, ask for a theme or clearly state the auto-selected theme

Step Override

Treat --step N as an explicit routing and pause directive.

• --step 3 means generate topic candidates and stop
• --step 3.5 means generate framework proposals and stop
• --step 6 means plan visuals and stop
• --step 7 means start from formatting and publishing
• --step 8 means start from post-publish maintenance tasks

Critical Quality Rules

1. Read references/writing-guide.md before drafting.
2. Before Step 4 drafting, gather fresh related information for the article subject. Do not draft “latest / trend / current state” articles from memory only.
3. In Step 3.5, choose both framework and article_archetype, then bind output_shape.
4. Keep the H1 title concise and WeChat-friendly.
5. The digest must not repeat the title.
6. Respect the blacklist and tone fields in {runtime_root}/clients/{client}/style.yaml.
7. Route from Step 1 based on the actual input type instead of forcing the full pipeline every time.
8. Respect explicit --step routing and stop at that step.
9. Step 5 is two-stage: 5A auto-fix then 5B editorial QA.
10. Step 5B must write quality-report.md beside the article bundle.
11. Before generating visuals, ask about image scope, style, and inline image density unless the user already specified them.
12. In Step 6, the agent must choose the explicit cover type and explicit inline targets before calling toolkit commands. Prefer paragraph/content-block anchors; use H2/H3 headings only as fallback. Do not let the toolkit infer them from article content.
13. Before preview or publish, ask for a theme or explicitly state the chosen theme.
14. When the workflow reaches publishing, publish directly to WeChat drafts. Do not ask for an extra publish confirmation.
15. If publishing fails, fall back to local preview instead of stopping the workflow.
16. Use configured image providers only: Gemini, OpenAI, Doubao, or Qwen.

Pipeline Overview

The workflow is organized as follows:

1. Load client configuration and identify the input type
2. Fetch topical signals when no concrete topic is provided
3. Pick a topic angle
4. Pick an article framework and article archetype
5. Draft the article with archetype-bound writing rules
6. Run 5A auto-fix and 5B editorial QA
7. Ask about visuals and generate optional assets
8. Format, preview, and publish to WeChat drafts
9. Update history, analytics, and lessons

The later stages also cover corpus ingestion and playbook refresh when the user asks to feed source material or update the client's writing system.

Resilience: Never Stop on a Single-Step Failure

Do not stop the entire workflow because one step fails.

Follow these rules:

• If a step fails, try its documented fallback first
• If publishing fails, fall back to local preview
• If image generation fails, continue with a text-only or prompt-only version
• If hotspot fetching fails, ask the user for a direct topic and continue
• If fresh-info search fails during drafting, tell the user what is missing and either continue with explicit freshness limits or ask for source links
• Stop only when required input is missing or both the primary path and fallback fail

The goal is to deliver the best usable outcome, not to terminate the whole workflow on a local failure.

Operations

Operational tasks for this skill include:

• First-run setup
• Client onboarding
• Corpus ingestion and playbook refresh
• Analytics backfill
• Edit learning and summarization

See references/operations.md for detailed procedures.

If {runtime_root}/clients/{client}/ does not exist, do not silently invent a full client profile. Follow the onboarding guidance and template references defined in the operational docs.

Gotchas — Common Failure Patterns

Common failure patterns include:

• WeChat publishing fails because the current public IP is not whitelisted
• Local preview looks correct but WeChat rendering still breaks due to unsupported CSS
• Edit learning is run on mismatched draft and final files
• corpus/ is treated as an automatic archive instead of a curated reference set
• Toolkit source changes are made without rebuilding dist/
• Topic selection relies on script output without real client-fit judgment

When these happen, check configuration, file paths, client state, and build artifacts first.

Comparison

Compared with a generic "write me a WeChat article" prompt, this skill:

• Covers the full path from topic to WeChat drafts
• Handles formatting, SEO polish, visuals, and publication together
• Supports client-level history, style memory, and playbook maintenance
• Enables a repeatable workflow instead of a one-off text generation task

Use it when the goal is not just to generate an article, but to operate a repeatable WeChat publishing workflow.

References

• Execution detail: references/pipeline.md
• Operations and onboarding: references/operations.md
• Writing quality: references/writing-guide.md
• Archetype routing: references/article-archetypes.md
• Editorial QA: references/editorial-qa.md
• Rewrite examples: references/rewrite-examples.md
• Framework selection: references/frameworks.md
• Topic selection: references/topic-selection.md
• SEO rules: references/seo-rules.md
• Visual prompts: references/visual-prompts.md
• Theme selection: references/theme-selection.md
• WeChat constraints: references/wechat-constraints.md
• CLI reference: references/cli-reference.md