WeChat Official Account Publishing Skill v3.0.0#

Language#

Always match the user's language. Chinese input → respond in Chinese. English input → respond in English.

Core Principle#

This skill is an editor-first, self-improving publishing pipeline — not a raw uploader.

Priority order:

1. Humanized writing quality
2. WeChat compatibility
3. Visual readability
4. Publishing speed

Never sacrifice compatibility for fancy formatting. Never sacrifice readability for automation.

---

Quick Reference#

ModeWhenKey flagStandardDefault—TurboContent already polished--turbo or turbo: true in frontmatterRawSkip all editorial pass--rawImage-textShort post with imagesauto-detected

Supported tones: 专业评论 · 行业快评 · 知识科普 · 观点专栏 · 深度分析 · 快讯播报

Reference files (load when needed):

• references/fsm.md — Full FSM state table
• references/quality-scoring.md — Scoring rubric details
• references/editorial-rules.md — Humanization + formatting rules
• references/config-reference.md — All EXTEND.md keys + full example
• references/scripts.md — Script CLI reference
• references/troubleshooting.md — Error levels + fix guide

---

FSM Workflow (Summary)#

Load references/fsm.md for the complete state transition table.

State sequence:

text

idle
  → loading_config
    → resolving_account
      → classifying_input
        → [turbo?] editorial_pass_light / editorial_pass_full
          → metadata_validation
            → packaging_check
              → pre_publish_confirm (if enabled)
                → publishing
                  → reporting → idle

Key transition rules:

• editorial_pass_full loops max 2 iterations if quality_score < threshold; then warn + continue
• packaging_check auto-repairs what it can; halts only for missing cover (API mode)
• publishing auto-retries on LEVEL 1 errors (max 2 retries); escalates on LEVEL 3+
• Any state can transition to error_report on unrecoverable failure

---

Error Levels#

LevelTypeBehavior1Transient (token expiry, network blip)Auto-retry silently (max 2, wait 3–10s)2Auto-repairable (missing summary, cover fallback, over-length)Repair + flag in confirmation3Requires user action (credentials missing, empty content)Halt + exact setup guide4Unrecoverable API errorLog + show error code + link to mp.weixin.qq.com

See references/troubleshooting.md for full repair paths.

---

Quality Scoring (Pre-publish)#

Score is computed before pre_publish_confirm. Load references/quality-scoring.md for rubric.

DimensionMaxSignalopening_hook20First paragraph creates pull or tensionheading_quality15Reader-question style, not generic labelsparagraph_rhythm15Varied length, no mechanical same-beat paragraphsai_tone_density20Inverse: fewer blocked phrases = higher scoreterm_preservation10All protected_terms intactending_quality10Human close + CTA presentlength_fit10Within default_article_length, not paddedTotal****100

Thresholds:

• ≥ 70 → proceed normally
• 60–69 → proceed with warning in confirmation
• < 60 → another editorial pass (max 2 total)
• < 50 → halt even in Turbo mode

Show score + brief per-dimension note in pre_publish_confirm.

---

Turbo Mode#

Activated by --turbo flag or turbo: true in frontmatter.

Skips: deep humanization pass, pre-publish confirmation (auto-proceed), quality score display (only blocks if < 50)

Still runs: metadata resolution, packaging check (auto-repair), cover resolution, token cache check, publish, log

Use when content is already manually polished, or when batch-publishing multiple articles.

---

Editorial Pass (Standard)#

Load references/editorial-rules.md for full humanization rules, AI-tone signal list, and formatting downgrade table.

Parallelizable side-computations (run while main humanization pass runs):

• Slug generation (from title)
• Tail keyword generation (from article topic)
• Profile block inference (from account config)
• Image count assessment (from article length)

Humanization strength:

• light: minimal cleanup, preserve style
• medium: default; improve rhythm, headings, opening, ending
• strong: substantial rewrite while preserving meaning

Industry terminology protection: Never simplify or paraphrase protected_terms. These are load-bearing words for the target audience.

---

Image Handling#

Count guidance by length:

Article lengthPreferred images< 800 chars2800–1500 chars2–31500–2000 chars3 (optionally 4)> 2000 charscompress content first, then reassess

Source resolution order:

1. Inline images already in markdown / HTML
2. imgs/ directory beside article file
3. User-supplied via --images
4. If insufficient: warn in confirmation, do not block publish

---

Metadata & Packaging#

Metadata resolution (each field):

• Title: CLI → frontmatter/H1 → strongest H2 → first meaningful sentence → auto-generate
• Summary: CLI → frontmatter → first paragraph trimmed → auto-generate
• Author: CLI → frontmatter → account default → global default

Packaging checklist (auto-repair if missing, unless --raw):

• Length within default_article_length
• 2–4 images (warn if under; do not block)
• Long-tail keyword block (3–8 phrases, topic-specific)
• Public-account profile block

Cover image resolution (API mode only):

1. --cover → frontmatter cover fields → imgs/cover.png → first inline image → halt if still missing

---

Access Token Cache#

Cache location (in priority order): amiao/.wechat-token-cache → ~/amiao/.wechat-token-cache

Behavior: read cache → verify expiry → refresh if expired/invalid → update cache. Failure to refresh = LEVEL 1 retry, then LEVEL 3 if persistent.

---

Publish Log + Auto-tune#

Every successful publish appends one record to amiao/.publish-log.yaml:

yaml

- date: <ISO8601+08:00>
  account: <alias>
  article_title: <title>
  quality_score: <0-100>
  humanize_level: <light|medium|strong>
  tone: <tone>
  article_length: <chars>
  image_count: <n>
  publish_method: <api|browser>
  media_id: <id>
  issues_found: [<list of flags>]
  cta_type: <type>
  tail_keywords: [<list>]

Auto-tune (triggered after every 10th publish cycle):

• Avg quality_score < 72 → suggest reviewing humanization defaults
• Tail keywords repeat too uniformly → suggest refreshing default_tail_keywords
• Article length consistently under/over → suggest adjusting default_article_length
• One tone consistently scores higher → suggest making it default_tone

Report suggestions at end of 10th cycle. Never auto-apply; always surface to user.

---

Pre-publish Confirmation Summary#

Show when confirm_before_publish: true (default), or when packaging was repaired, or when image count is below target.

text

───────────────────────────────────────
发布前确认 / Pre-publish Summary
───────────────────────────────────────
账号 Account   : [name]
方式 Method    : [api / browser]
主题 Theme     : [theme] [color]
编辑 Editorial : [humanize level] · [tone]

标题 Title     : [title]
摘要 Summary   : [summary]
字数 Length    : ~[N] 字
图片 Images    : [N] 张 [⚠ below target if applicable]
质量评分 Score : [N]/100  ([brief note if < 70])

封面 Cover     : [resolved / ⚠ missing]
长尾词 Keywords: ✓ present / ⚠ missing
账号介绍 Profile: ✓ present / ⚠ missing

[⚠ Any auto-repairs applied]
───────────────────────────────────────
确认发布？(y/n / 输入 e 返回编辑)

If confirm_before_publish: false and no warnings: skip confirmation, show summary inline in completion report instead.

---

Completion Report#

text

✓ WeChat Publishing Complete!

Input    : [type] · [path]
Method   : [API / Browser]
Account  : [name]
Theme    : [theme] [color]
Editorial: [humanize] · [tone]

Article:
  Title   : [title]
  Summary : [summary]
  Images  : [N]
  Comments: [open/closed] · [fans-only/all]
  Score   : [N]/100

Result:
  ✓ Draft saved to WeChat Official Account
  media_id: [id]   (API mode)

Files:
  [• slug.md if created from plain text]

Next: https://mp.weixin.qq.com → 内容管理 → 草稿箱

---

Scripts#

Load references/scripts.md for full CLI reference.

ScriptPurposescripts/wechat-api.tsArticle via APIscripts/wechat-article.tsArticle via browserscripts/wechat-browser.tsImage-text postscripts/md-to-wechat.tsMarkdown → WeChat HTMLscripts/check-permissions.tsEnvironment preflight

Runtime: bun → npx -y bun → explain + suggest install

Critical: Publishing scripts handle markdown conversion internally. Never pre-convert markdown to HTML before calling the publish script.

---

Configuration#

Load references/config-reference.md for all supported EXTEND.md keys and a full annotated example.

EXTEND.md location (checked in order):

bash

amiao/amiao-post-to-wechat/EXTEND.md          # project-level
${XDG_CONFIG_HOME:-$HOME/.config}/amiao/amiao-post-to-wechat/EXTEND.md
$HOME/amiao/amiao-post-to-wechat/EXTEND.md    # user-level

If not found → run first-time setup ( references/config/first-time-setup.md) before proceeding.

Value priority: CLI → frontmatter → EXTEND (account → global) → skill defaults

---

OpenClaw / ClawHub Rules#

• SKILL.md is the primary agent contract; keep it self-contained and readable
• Relative paths assume scripts/references live beside this skill
• No hard-coded machine-specific absolute paths
• All defaults degrade gracefully when optional config is absent
• Skill name, version, and metadata must stay clean for ClawHub distribution
• Favor backward-compatible defaults on version updates