A content farm with a conscience: agent skills that refuse to auto-post
Ten modular agent skills that run the whole self-media pipeline — brief to analytics — while forcing humans to approve every publish.

What it does
This is a suite of ten agent skills (Claude Code plugin, or installable via the Vercel skills CLI for Codex/Cursor) covering the full self-media content loop: turning a vague idea into a brief, account strategy, trend and competitor research, platform-native copy, short-video planning (including digital-human and animated-explainer production), publishing to WeChat/YouTube/Douyin/Xiaohongshu, and post-publish analytics. A controller skill routes and tracks state; nine single-responsibility modules do the actual work.
The interesting bit
The design philosophy is unusually cautious for a content-automation tool. There are five mandatory human confirmation gates — direction, platform, headline, final draft, and publish authorization — and the skills never auto-mass-post. It’s also tool-agnostic: collection and publishing are runtime adapter layers that discover whatever capabilities your environment actually has, rather than hardcoding vendors.
Key highlights
- One shared fact/evidence base per topic, but titles, hooks, structure, and calls-to-action are rewritten per platform instead of cross-posted verbatim
- Eight built-in illustration style presets with platform routing; style is chosen before images are generated, and account preferences override defaults
- Strict evidence rules: key numbers need sources, and facts, judgments, inferences, and recommendations are kept distinct — no fabricated data or testimonials
- Security boundaries: no cookies/tokens/secrets in task cards or logs, no auto-liking/following/DMing, and no scraping competitors via your main account’s session
- Digital-human avatars and voice assets are uploaded by the user on the provider’s platform; the repo only indexes delivered clips and final videos, never raw biometric material
- A dependency-free
validate.pychecks structure, frontmatter, and links; CI runs it plus a real install test on every commit
Caveats
- Platform rules and limits change; the README itself says its specs are point-in-time experience and you should defer to official docs and the actual publish UI
- Real account operations (WeChat draft writing) carry risk — the disclaimer puts rate-limiting, bans, and content-liability squarely on you
- The README is mostly Chinese; there’s an English version, but the ecosystem it targets (WeChat, Xiaohongshu, Douyin) is decidedly China-centric
Verdict
Worth a look if you produce content across Chinese social platforms and want an agent that drafts aggressively but publishes only when you say so. If your stack is Twitter/Substack, the platform modules won’t map — though the brief/strategy/analytics skills are portable in principle.
Frequently asked
- What is yanhua1010/self-media-content-workflow?
- Ten modular agent skills that run the whole self-media pipeline — brief to analytics — while forcing humans to approve every publish.
- Is self-media-content-workflow open source?
- Yes — yanhua1010/self-media-content-workflow is open source, released under the MIT license.
- What language is self-media-content-workflow written in?
- yanhua1010/self-media-content-workflow is primarily written in JavaScript.
- How popular is self-media-content-workflow?
- yanhua1010/self-media-content-workflow has 718 stars on GitHub and is currently cooling off.
- Where can I find self-media-content-workflow?
- yanhua1010/self-media-content-workflow is on GitHub at https://github.com/yanhua1010/self-media-content-workflow.