An AI Content Production Workflow Skill for Self-Media, With a Real Language Catch

Most attempts at giving an AI agent a content production workflow skill start from scratch: a single long prompt trying to cover strategy, writing, and publishing all at once. self-media-content-workflow, an open-source project by developer yanhua1010, takes a different shape. It splits the job into nine linked Claude Skills, each owning one stage of running a content channel, from positioning through to publishing, with human approval built into the process rather than bolted on. As of today, the repository has 368 stars and 50 forks on GitHub, MIT licensed, created in late July 2026.

What the skill suite actually does

The repo installs as a set of nine modular skills that route through one orchestrator:

• self-media-content-workflow: the router. Reads a request, decides which skill it needs, tracks state, and enforces approval checkpoints.
• self-media-content-brief: pins down audience, goal, evidence, angle, tone, and constraints before anything gets written.
• self-media-content-strategy: account positioning, content mix, series planning, topic pools, and the content calendar.
• self-media-trend-radar: trend tracking, keyword research, competitor teardowns, and original topic generation.
• self-media-platform-copywriting: native copy for X, Xiaohongshu, WeChat, and short-video platforms, plus a shared visual style library so the same core content doesn't look mismatched across channels.
• self-media-short-video: hooks, spoken script, storyboard, captions, and a shoot plan.
• self-media-content-analytics: data quality checks, comparable baselines, attribution, and experiment design.
• self-media-content-delivery: milestone files, versioning, path verification, and publishing packages.
• self-media-wechat-publisher: WeChat-specific formatting, image upload, draft creation, and image-message posts.

The design choice worth naming is the approval gates. The system builds toward five mandatory human checkpoints: direction approval, platform approval, title approval, final-copy approval, and publishing authorization. It creates drafts and packages, not live posts. There's no automated mass publishing built in, which matters for anyone who has watched an agent post something to a real account before a human looked at it.

Installing it

Two paths, both documented in the README. Inside Claude Code, it installs as a plugin:

/plugin marketplace add yanhua1010/self-media-content-workflow
/plugin install self-media-suite@self-media

For other agents, a Node.js-based skills CLI handles it:

npx skills add yanhua1010/self-media-content-workflow

Claude Code is the recommended path but not required; Node.js is only needed for the alternative installer.

The language reality, stated plainly

This is where the honesty has to be direct. The repo's main README.md is written in Chinese first. There is a README.en.md, and it is a genuinely thorough translation, not a token stub: it covers installation, the architecture, all nine skills, the approval model, and the safety disclaimers in full. Anyone deciding whether to try this tool can read the whole pitch in English.

But the README is not the whole product. The actual instructions each skill runs on, the SKILL.md files inside skills/self-media-content-strategy, skills/self-media-trend-radar, and the other seven directories, are written in Chinese only. There is no English version of the operative content. Claude itself reads Chinese without difficulty, so the skill will still function when invoked. What an English-speaking user loses is the ability to open a skill file, read exactly what instructions it's giving the model, and adjust or audit them without running a translation pass first. For a tool whose whole value proposition is a transparent, inspectable playbook, that's a real gap, not a cosmetic one.

There's a second layer to the language question beyond the files themselves: the platform focus. Two of the four platforms the copywriting skill targets by name, Xiaohongshu and WeChat, are Chinese platforms with no direct Western equivalent. X is covered, and "short-video platforms" is mentioned generically without naming which ones. A creator running channels entirely outside the Chinese platform ecosystem will find real value in the strategy, brief, trend-radar, and analytics skills, which are platform-agnostic, but will be building the platform-copywriting and WeChat-publisher pieces around use cases that don't map cleanly to their own channels.

Traction, and what it actually shows

368 stars and 50 forks, as of today, on a repository created July 22, 2026, with its last code push July 26. That's a genuinely new project, about three weeks old, not an established one with a track record. The star-to-fork ratio is worth a second look: a fork count that's roughly a seventh of the star count is unusually high for a project this young, and it suggests some of the people finding it are treating it as a starting template to customize rather than just a tool to install as-is, which tracks with how the skill suite is built, modular pieces meant to be adapted per account.

It's not a widely adopted, battle-tested system yet. It's a well-structured, actively maintained early-stage project (a v0.1.0 release, an open CI validation script, zero open issues at last check) from a single maintainer.

Who this is actually for

Best fit: a Claude Code user, comfortable reading or machine-translating Chinese source files when needed, who wants a structured, approval-gated workflow for running a content channel end to end, especially one that touches Xiaohongshu or WeChat directly, where this suite's platform-specific skills carry the most unique value. It's also a reasonable reference architecture to study even without adopting it whole: nine narrow skills routed through one orchestrator, with mandatory human checkpoints, is a pattern worth borrowing regardless of platform.

Weaker fit: anyone who wants to open the skill files and read, tweak, or extend the actual instructions without a translation step first, or anyone whose channels live entirely on English-language, non-Chinese platforms, since two of the nine skills and part of the copywriting skill are built around platforms they won't use. For that reader, the strategy, brief, trend-radar, short-video, and analytics skills still transfer, but roughly half the suite's platform-specific value doesn't.

The honest verdict

self-media-content-workflow is a real, working, MIT-licensed tool, not a thin or misleading repo. The English README is better than the "Chinese-first" label alone suggests. But the operative skill instructions stay in Chinese only, and the platform coverage leans toward Chinese social media by design. Combined with genuinely modest, three-week-old traction, this is worth trying for the architecture and for anyone already working Chinese platforms, and worth watching rather than adopting outright for everyone else.