The same thing needs a different "language" on different platforms —Each platform is a different room; when you enter, follow that room's rules first(For specific differences, see "Platform Tone Quick Reference" below). This skill turns "same material → native copy for each platform" rewriting into an executable process.
This skill only generates copy, does not publish, does not call APIs, does not auto-post. Output = post text for each platform that can be directly copied and pasted (+ image/tag suggestions).
Material to promote
$ARGUMENTS
Core commandments (four)
- Give value first, then talk about yourself. The essence of anti-hard-advertising in any community is "are you taking (asking for follows/clicks) or giving (substantive content/inspiration/solutions)". One rule that works across all platforms: the body of the post is something readers can take away; promotion is a byproduct.
- Follow the room's rules. Word count, titles, tags, tone, and CTA forms differ by platform; copying one draft and posting it everywhere = it sounds native nowhere.
- Don't invent algorithm folklore. Claims like "external links get 50% downranking", "more than 3 hashtags triggers spam", "replies carry 27x weight" are mostly inferences from third-party blogs and unconfirmed by the platforms officially (X's official open-source algorithm has already removed manual features).Write these as "practical suggestions", never assert them as platform iron laws. The only things you can assert are what's officially stated (character limits, explicit spam policies).
- De-AIGC-ify it, make it sound like a real person. Every piece must pass an AI-tone self-check — delete formulaic connectives (firstly/secondly/in summary/In conclusion), break up uniform sentence lengths, inject real details only a real person would know (numbers/pitfalls/scenarios), and adopt the platform's dialect. Most platforms have already cracked down on hollow AI content; human flavor simultaneously passes AI detection, platform algorithms, and readers.The most reliable way to remove AI flavor is to inject real information, not to rely on templates. Don't promise "guaranteed to pass detection" (detection is probabilistic). See
references/anti-ai-detection.md。
First action: Extract the material's core (shared across all platforms)
Before writing for any platform, first squeeze the material into these 5 slots — this is the raw material for all subsequent rewrites:
| Slot | Question | Example |
|---|---|---|
| One-sentence value | Who it helps, what it solves, why it works | "A skill that lets Claude Code rewrite and promote copy across multiple platforms" |
| Target audience | Who will care, in what scenario | Indie developers / content creators |
| Credible evidence | demo / data / screenshots / comparison / real experience | GIF, star count, before-after |
| CTA goal | What you want readers to do (single) | Try it / star / bookmark / discuss / follow |
| Taboos | What cannot be said (exaggeration / prohibited words / traffic-driving red lines) | Advertising law superlative terms, unverified efficacy-style claims |
🔴 CHECKPOINT (before rewriting): All 5 slots must be filled, especially "credible evidence"—promotion without evidence will be treated as a soft ad in any technical community, andreal details are also the core raw material for removing the AI tone(AI cannot make up private numbers/pitfalls). If evidence is missing → first ask the user for it (screenshots/data/real usage stories), do not fabricate it on the user's behalf.
Workflow A: One-click full set for multiple platforms (in order)
| Step | Action | Completion criteria |
|---|---|---|
| 1 | Extract the core(the 5 slots above) | All 5 slots have been filled (especially credible evidence is non-empty) |
| 2 | Choose platforms: If the user did not specify → recommend based on the nature of the material (open-source tools → dev camp; consumer/lifestyle-oriented → recommendation camp; if it touches both → produce for both sides) | The list of platforms to publish on this round has been listed, and the user has confirmed it or it has been determined based on the nature of the material |
| 3 | Adapt per platform: For each platform, check the corresponding row in "Platform Tone Quick Reference" + read references/ detailed profile, and produce one piece for each |
Each selected platform has produced one draft body |
| 4 | Self-check each piece: Go through that platform's taboos + the four core commandments (anti-hard-sell) + AI-tone removal self-check(references/anti-ai-detection.md); whether the title/word count/tags are compliant |
Each piece has gone through that platform's taboo list + AI-tone removal self-check checklist item by item, with no omissions |
| 5 | Package and deliver: Each piece is labeled "Platform / Title / Body / Tags / Image suggestions / CTA", ready for the user to copy and use | Each piece has all six elements complete, and the user can copy directly without asking again |
Workflow B: Single-platform refinement (in order)
- Confirm the platform + read
references/that platform's detailed profile. Completion criteria: the target platform is determined, and the corresponding detailed profile has been read. - Extract the core (if not already extracted). Completion criteria: all 5 slots have been filled.
- According to that platform'stitle formula + opening hook + body structure + CTA formpolish paragraph by paragraph (for the method, see
references/playbook.md). Completion criteria: the title/opening/body/CTA sections have all been written according to that platform's formula. - Go through that platform's number-one taboo + prohibited words (for recommendation/WeChat public accounts especially check Advertising Law absolute terms) + AI-tone removal self-check. Completion criteria: the taboo list + AI-tone removal checklist have been gone through item by item, with no omissions.
- Provide 2-3 alternative titles (the title determines the open rate). Completion criteria: 2-3 optional titles have been provided for the user to choose from.
Platform Tone Quick Reference
Word counts are "approximate values/community conventions"; only the X character limit is officially specified (280 / Premium 25000). The rest are subject to the platform's current actual practice.
Developer / technical community camp (tone: sincere, verifiable, low marketing flavor, show-don't-tell)
| Platform | Audience/format | Volume | Title/opening | Tags | Appropriate CTA form | Number-one taboo |
|---|---|---|---|---|---|---|
| GitHub | Engineers evaluating code (README/Release/Discussion) | The first screen of the README decides life or death | One-sentence value + demo GIF | Selected badges (version/build/license) | "Star if useful""Quick start""Contributions welcome" | Star farming / rewarding star exchanges / posting ads in other people's issues (violates the AUP) |
| linux.do | Chinese developer forum (Discourse, trust-level system) | Medium-to-long post, body-focused | Plain and straightforward, no clickbait | Choose the right category node | Sincere sharing + invite discussion | Pure hard advertising / begging-style traffic driving → downvoted and banned |
| Twitter/X | English tech circle | Single post 280 characters(Premium 25000) | First post = hook (bold claim/specific promise), no links | 0-1 hashtag | Put external links inreply #2; end with a call for replies | First post stuffed with links + many hashtags; marketing tone |
| Hacker News | Extremely picky technical readers (Show HN) | Title + short body | Show HN: 工具名 – 一句话做什么(no clickbait) |
No tag system | Honestly say in the post what was done, ask for feedback | Clickbait / marketing tone / exaggeration |
| Niche subreddit | Depends on the sub | Specific, not clickbait | Follow subreddit rules | Contribute first, then occasionally self-promote (9:1 self-promotion unwritten rule) | Ignoring sub rules and spamming self-promotion → ban | |
| V2EX | Chinese geeks | Medium-short post | Plain | Post in the right node (e.g. "Share & Create") | Share-and-create mindset | Wrong node / hard advertising traffic driving |
| Juejin | Chinese technical writing community | Long technical article | Substance-style title | Technical tags | Lightly mention the project at the end | Whole piece is advertorial with no substance |
Mass recommendation / Q&A camp (tone: colloquial, empathetic, altruistic, scenario-based)
| Platform | Audience/format | Volume | Title/opening | Tags | Appropriate CTA form | Number one taboo |
|---|---|---|---|---|---|---|
| Xiaohongshu | Find life/consumption inspiration (notes) | Title about ≤20 characters, short body, paragraphs, emoji | Pain point/benefit hook + emoji, title determines opens | Several topic # (vertically relevant) | "Sisters go for it" "Save this" "Ask me in the comments" | Advertising law superlatives (best/first/national-level), hard traffic-driving external links → throttled |
| Zhihu | Seek professional answers (answers/articles/thoughts) | Long-form substance, Q&A style | Give the conclusion first/counterintuitive hook | Topic tags | Disclose conflicts of interest + naturally bring it out in the substance | Advertorial with no substance / being clever / not disclosing conflicts of interest |
| WeChat Official Account | Subscribers (articles) | Long article, heavy formatting | Title determines open rate (restrained faction) | No public tags | Guide at the end (Wow/Share/Read Original) | Inducing sharing (platform restriction), excessive clickbait |
| Jike | Light social updates (circles) | Short update | Relaxed, sincere, colloquial | Post in the right circle | Friend-recommendation tone | Ad-speak tone / hard-sell voice |
For detailed writing guidance per platform (opening hook templates, body skeleton, CTA tiers, banned word list), see references/platforms-dev.md and references/platforms-social.md。
The fundamental differences between the two camps (choosing the wrong tone = all ruined)
| Dimension | Developer/tech community | Mass-market种草/Q&A |
|---|---|---|
| What readers want | Verifiable technical value, runnable code | Emotional resonance, life solutions, inspiration |
| Source of trust | demo / data / open source / honesty | Real experience / persona / detailed scenarios |
| Tolerance for marketing flavor | Extremely low (salesy = points deducted) | Medium (but must be "soft"; superlative claims are illegal) |
| Title | Specific, not clickbait | Hook, pain point, benefit up front |
| Most fatal mistake | Exaggeration + marketing tone | Dryly listing specs, sounding like an ad |
New platform profile template (self-extending)
When the user wants a platform not in the quick-reference table (such as ProductHunt / Bilibili posts / Maimai), first have them fill in this profile, then rewrite accordingly:
平台名:
受众画像(谁、什么场景):
内容形态(帖/文/动态/视频文案):
体量约束(标题字数 / 正文长短 / 标签数):
调性(技术真诚 / 种草共情 / 专业问答 / 轻社交):
标题与开头偏好:
CTA 得体形态:
头号禁忌(会被折叠/删/反感的):
违禁词风险(是否受广告法/平台敏感词约束):
See details in references/new-platform-template.md。
Failure modes (trigger → front-line fix → fallback if still failing)
| Trigger | Front-line fix | Fallback if still failing |
|---|---|---|
| Material lacks credible evidence (no demo/data/screenshot) | 🔴 Stop, ask the user for real evidence (screenshots/numbers/usage stories) | User cannot provide it → write only "the verifiable parts," leave blanks where evidence is missing, never fabricate data/positive reviews |
| Copy is judged by the platform as soft advertising / collapsed | Cut promotional claims, change the main body to takeaways/inspiration readers can take away, reduce promotion to a byproduct | Still collapsed → for that platform, switch to a pure value post, move the CTA to comments/end of post |
| 种草/WeChat Official Account mistakenly uses Advertising Law superlative terms (most/first/national-level), triggering throttling | Switch to experiential phrasing ("it felt handy when I used it"), delete absolute terms | Unsure whether it is prohibited → mark "suspected superlative term, needs re-check," do not force-publish |
| After removing AI tone, still suspected of being flagged by AI detection | Add real private details (numbers/pitfalls/scenarios) + break up uniform sentence length, do not rely on typos | Detection is probabilistic → tell the user "features have been reduced but passing is not guaranteed," do not promise |
| Platform rules/word counts/review have changed (quick-reference table is outdated) | Take the platform's current actual conditions as authoritative, recalibrate constraints according to official explicit text | Cannot verify current rules → mark "rules may have changed, re-calibrate against the target platform before posting" |
Scope of application
- This skill coverscopy generation, but does not cover publishing, does not cover paid traffic, and does not guarantee conversion or traffic (platform algorithms are uncontrollable and mostly black boxes).
- Platform rules/word counts/reviewchange quickly: only the X character limits (280/25000) and the GitHub AUP red lines are official explicit text verified in this round (2026-06); all other volume/tag matters are community conventions, subject to the platform's current actual conditions.
- Timing/operations techniques for "how to post so it goes more viral" are not the focus of this skill (the user only needs copy); the quick-reference table only retains constraints that affectcopy writing.
Reference set (read as needed)
| File | Purpose |
|---|---|
references/platforms-dev.md |
Developer camp detailed file: opening hooks, body skeletons, CTAs, and taboos for GitHub/linux.do/Twitter-X/HN/Reddit/V2EX/Juejin (including GitHub AUP red lines and the original text of X character rules) |
references/platforms-social.md |
Recommendation Q&A camp detailed file: title formulas, body structures, prohibited word lists, and conflict-of-interest disclosures for Xiaohongshu/Zhihu/WeChat Official Accounts/Jike |
references/playbook.md |
General copywriting methodology: core appeal (PAS/AIDA), title formulas, opening hook routines, three-tier CTAs, rewriting process, and the advertising law absolute terms list |
references/anti-ai-detection.md |
De-AIGC-ification: Chinese and English AI-tone marker words, 7 de-AI techniques, post-generation self-check list, relationship with compliance, and honest boundaries |
references/new-platform-template.md |
New platform profile fill-in template + method for deriving tone from the profile |