// the find
alchaincyf/huashu-skills
花叔全部开源 Agent Skills 总目录：16 旗舰 + 14 人物视角 + 22 内置共 52 个 skill，分层分类 + AI Agent 安装协议 + 机器可读 skills.json + 更新检查机制
A master index for Huashu's open-source Agent Skills, covering flagship skills in standalone repos, persona skills distilled with nuwa-skill, and 22 lightweight skills bundled in this repo. It is mostly a README plus a skills.json catalog, so the actual skill code lives elsewhere. It is useful for someone already following this author, or for an agent that needs a machine-readable map of the ecosystem.
The routing table maps tasks to skills ("I want X, use Y"), which is more useful than a bare list, and skills.json gives agents the same mapping in a parseable form. The huashu-design name collision between the standalone repo and the bundled lite version is disclosed up front, with a concrete rename procedure in the agent protocol instead of leaving users to hit it. Bundled skills follow a consistent layout with SKILL.md plus references/ and scripts/ where needed, and the update checker is an actual script (check_updates.py) rather than a promise.
The counts don't agree with each other: the intro says 53, the English summary says 52 with 16 flagship, and the tables list 18 flagship, which sums to 54. For a repo whose whole job is cataloguing, that is the first thing a reader will check. The update mechanism is the weakest part: installed skills are told to check GitHub every 30 days and write dotfiles into ~/.claude/skills, while installs are unpinned clones of default-branch HEAD with no tags, checksums, or signatures, so adopters are trusting whatever the author last pushed. The README says these skills work across Claude Code, Codex, and Kimi Code, but the catalog describes at least two as 豆包-only (huashu-icon-set and huashu-slide-doubao), and the visible tree has no test files for any bundled skill, so the quality claims rest on the README alone.