• User says "帮我推广这些 skills" / "设置每天自动发推广" / "研究一下别人怎么推广 skill 的"
• User explicitly calls

  /wjs-promoting-skills

• User asks "今天准备推哪个 skill" / "outbox 里有什么"

/publish-skill <name>

~/.claude/skills/wjs-promoting-skills/setup.sh

1. Check prerequisites:

   claude

   CLI,

   xurl

   (and

   xurl whoami

   returns the user),

   jq

2. Run

   research-marketplaces.sh

   once to generate initial

   state/research.md

   (this is the only time it will actually research online, it will auto-refresh once a month next time)
3. Render

   com.jianshuo.wjs-promoting-skills.plist.template

   into a real plist and place it in

   ~/Library/LaunchAgents/

   , then run

   launchctl bootstrap

~/.claude/skills/wjs-promoting-skills/uninstall.sh

04:00 →
  Step 1: list-skills.sh        → List of all current wjs-* skills + their last promotion time
  Step 2: pick-next-skill.sh    → Select today's skill according to rules (see §Rotation Rules)
  Step 3: Check skip conditions → Promoted within 7 days / SKILL.md unchanged / Today's content already in outbox → exit 0 directly
  Step 4: make-plan.sh <skill>  → If the skill has no plan or the plan is older than 30 days, regenerate it
  Step 5: Run daily-post.md via claude -p
            └── Input: SKILL.md + plan + research.md + last 30 days of history.jsonl
            └── Output: An X text ≤ 280 characters (including repo URL)
  Step 6: xurl -X POST -d '{"text": "..."}' /2/tweets   ← Publish directly
  Step 7: Run community-drafts.md via claude -p
            └── Output: outbox/<date>/reddit-r-ClaudeAI.md / hn-show.md / discord-anthropic.md
  Step 8: Append one line to history.jsonl, and archive the posted X tweet to outbox/<date>/x-posted.txt

• research-marketplaces.sh

  refreshes

  state/research.md

  (check changes in copywriting of top skills on marketplaces that week)

• Re-run

  make-plan.sh

  for all skills with plans older than 30 days (rotate angles)

1. Skills never promoted before — Prioritize these to get new skills exposure early
2. Skills not promoted in the last 7 days — Among these, select those with recently modified SKILL.md (indicating the user just iterated on it)
3. Others — Select the skill that was promoted the longest ago among all skills

• ≤ 280 characters (X counts URLs as 23 characters, reserve 25)
• First line: skill name + one-sentence value proposition
• Middle: 2–3 specific capabilities / differences from the latest update
• Second-to-last line: repo URL on a separate line (X will automatically render a preview card)
• At most 1 hashtag (

  #ClaudeCode

  or

  #ClaudeSkills

  )
• Prohibited: Marketing jargon, rocket emojis,

  AI-powered

  ,

  game-changer

  , exaggerated words
• Required: Match the user's tone (refer to the style of wjs-* skills in README.md: practical, specific, no hype)

• Angle A: Specific problem it solves (details like "75-minute 4K multi-camera 60GB re-encoding sync doubles disk usage")
• Angle B: Counterintuitive design decisions (like "sidecar over re-encode")
• Angle C: Workflow integration (what results can be achieved with other skills)
• Angle D: Differences from the latest update (if SKILL.md has changed since last promotion, focus on new features from the diff)

make-plan.sh

outbox/<date>/

• reddit-r-ClaudeAI.md

  : Long-form format, start with a hook, body covers scenarios + code snippets + links, end with an open question (Reddit prefers discussions over ads). Selfpost only, never post pure links

• hn-show.md

  : Show HN format, title

  Show HN: <skill> – <one line value>

  , body < 600 words, note open source license + repo URL, don't sell on HN

• discord-anthropic.md

  : Short post, 2–3 sentences + link, suitable for #show-and-tell channels

• wechat-followup.md

  (Optional): If the skill is suitable for a detailed WeChat official account article, provide a 200-word outline + several angles for direct use when the user writes the article later

• claude

  CLI is in

  $PATH

  , and

  claude -p "hello"

  runs successfully (headless mode)

• xurl

  is installed,

  xurl whoami

  returns

  jianshuo

• jq

  is installed (

  brew install jq

  )
• Public repositories: Each

  wjs-*

  skill has a GitHub repo, and

  git config --get remote.origin.url

  can retrieve the URL. This repository

  ~/code/claude-skills

  already meets this requirement
• macOS (uses

  launchd

  for scheduling; the Linux equivalent is systemd timer, but this skill doesn't implement it)

• Not "let AI run your account for you" — every post is based on your written SKILL.md, AI only reframes it from different angles
• Not a "cross-platform cross-poster" — only X supports direct publishing, others are drafts
• Not "fully autonomous" — you can stop it anytime with uninstall.sh, or use

  DRY_RUN=1

  to see what will be posted next time
• Not a replacement for

  /publish-skill

  — that's the manual single-post version, this is the daily automated version, they can coexist

• setup.sh

  runs successfully,

  launchctl list | grep wjs-promoting-skills

  shows it

• state/research.md

  exists and is > 1KB (indicates research has run)

• DRY_RUN=1 daily.sh

  runs completely without posting to X
• 4 markdown drafts exist in

  outbox/<today>/

• A new line is added to

  history.jsonl

  on the day of actual posting

Auto-publish: This skill is automatically synced to github.com/jianshuo/claude-skills via

~/.claude/skills-publish-hook.sh

(auto commit + push after each edit).

state/

and

outbox/

are excluded via the directory's

.gitignore

+ subdirectory

.gitignore

, and won't be pushed along with SKILL.md.