qiaomu-meta-skill
A meta-skill that turns a workflow into an agent skill: prior-art search, trigger evaluation and publication to GitHub
- 390 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- When publishing, the skill creates branches and pull requests in your name, merges them and cuts a GitHub release through your `gh` login
- For prior-art search it calls skills.sh, SkillsMP and GitHub, and for checks it runs its own Python scripts
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/qiaomu-meta-skillDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- When publishing, the skill creates branches and pull requests in your name, merges them and cuts a GitHub release through your `gh` login
- For prior-art search it calls skills.sh, SkillsMP and GitHub, and for checks it runs its own Python scripts
This is third-party code. Review the repository files before installing.
What it does
qiaomu-meta-skill walks an agent from raw material to a finished, installable skill. The source can be a prompt, an SOP, a chat log, an old skill, a script or a vague idea. The author describes a chain of steps: clarify the goal, find similar skills, design the package, test triggering, write the README, scan for leaked keys, open a pull request, cut a release and verify installation through `npx`. Before creating a new skill, the agent searches the skills.sh and SkillsMP catalogs, then returns to GitHub and checks the source, maintenance and license. For each candidate it records a decision: keep, adapt, reject or invent, so the result is not a patchwork of other people's text. Quality is tested on example phrases: which should call the skill, which should not and which sit on the border. Depth depends on the mode. A personal experiment follows the light Scaffold route, and a public skill goes through the full Governed set of checks, including permissions, rollback, secrets and claim checks. Publication happens only on a direct request and always through a branch and a pull request, never by writing straight to the default branch.
Who it is for. For people who build skills for Claude Code and Codex and for teams that need to turn a repeated process into a tested and published skill.
Good fit when
- You want to turn a long prompt or SOP into a reusable skill for a team
- You want an interview, one question at a time, to pull out an implicit working method as a skill
- Before writing your own skill you want to study and compare popular alternatives and avoid their mistakes
- A skill fires in the wrong place or never fires, and its description needs sharpening on real phrases
- A finished skill must be published on GitHub: README, license, secret check, pull request, release and a clean-environment install
Not a fit when
- You need a one-off summary, translation or ordinary documentation: the author explicitly excludes such tasks from skill cases
- You want to publish a package outside the skill format, such as an npm library: publication is built for skills only
- You design the skill with no access to GitHub and the catalogs: prior-art search and releases need network access and a `gh` login
- You want a skill in a neutral general style: by default the package is signed with the author's name and uses names with the qiaomu prefix
Example request
Take my release checklist and turn it into a skill: find similar ones first, then test triggering on examples and prepare the package for publication, but publish nothing.Limitations
It needs Node.js 18 or newer, npx and Python 3.9 or newer, and for publication also the GitHub CLI signed in, checked with `gh auth status`. Prior-art search needs access to skills.sh, SkillsMP and GitHub. If SkillsMP drops the connection or throttles requests, the researcher retries and marks the result `missing evidence` instead of inventing it.
The package is tuned to the author's style: by default names start with the qiaomu prefix and carry a signature with the author's contacts, so replace those defaults for your own use. The author himself advises forking and removing rules that do not fit you.
The README cites metrics: 35 of 35 unit tests, 23 of 23 trigger evaluations and 0 package validation problems. These are the author's measurements of a local version, and the author points to the Releases page for public confirmation. Part of the material is written in Chinese.
How to disable. Delete the qiaomu-meta-skill folder from ~/.agents/skills/ and from ~/.claude/skills/ if a link exists there. Skills created with it stay separate packages and can be deleted individually.
FAQ
How do I install qiaomu-meta-skill in Claude Code?
Run npx skills add joeseesun/qiaomu-meta-skill. If you want only this skill from the repository, add the parameter --skill qiaomu-meta-skill. You can verify the install with the validate_skill.py script from the skill folder.
How does qiaomu-meta-skill differ from an ordinary skill-creator?
It adds prior-art search in skills.sh and SkillsMP with a check on GitHub, a recorded decision for each candidate, trigger evaluation, secret scanning and self-contained publication through a branch, a pull request and a release with a clean-install check.
What are the Scaffold, Production, Library and Governed modes?
They are levels of strictness. Scaffold gives the minimum of files for a personal experiment. Production adds a README, an interface, trigger evaluation and install proof. Library and Governed suit shared infrastructure and public skills with permission, rollback and secret checks.
Does the skill publish anything to GitHub without asking?
No. Publication happens only on explicit request, and audit and evaluation requests stay read-only. Writing straight to the default branch is forbidden: everything goes through a branch, a pull request and a release.
How do I check a package before publishing?
Run python3 scripts/publish_skill.py with the skill path and the --dry-run flag: it changes nothing and writes nothing to GitHub. There are also validate_skill.py, trigger_eval.py and release_check.py for local checks.
Discussion
No questions or reviews yet.
Related
SkillsA meta-skill that creates, evaluates, packages and governs reusable agent skills
A skill that helps an agent author other skills to the agentskills.io spec: structure, metadata, progressive disclosure and validation
A meta-skill that takes a vague idea to a finished agent skill: it designs, writes the files, checks them and refines by evidence
A skill that designs, tests, evaluates and packages other agent skills, starting from architecture
skills (npx skills)
skills
A skill package manager: installs, finds, updates and removes SKILL.md packages from GitHub and other git hosts across many agents
Meta Skill (ms)
Meta Skill
Meta Skill (ms) keeps skills locally in SQLite and Git, searches them with a hybrid index and serves agents over MCP
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| qiaomu-meta-skill (this card) | no | 390 | |
| Yao Meta Skill | no | 2,719 | the method this skill grew from: Skill IR, evaluation and governance, without built-in prior-art search and publication |
| skill-creator | no | 2,287 | helps an author write skills to the agentskills.io spec but does not search for prior art or publish |
README in short
The README explains the motive: a skill description is easy to make too broad or too narrow, so search, evaluation and safe release are needed. It continues with a comparison to a plain SKILL.md generator, a list of 28 skills the author has made with 18 of them public, the structure of the resulting package, seven process steps, a table of common errors and a sources section that includes the skill-creator from Anthropic and OpenAI.
README badge
Original SKILL.md
--- name: qiaomu-meta-skill description: | Research, create, improve, migrate, evaluate, package, install-check, govern, and safely publish qiaomu-flavored agent skills from workflows, prompts, transcripts, docs, SOPs, runbooks, scripts, or notes. Use for new or existing skills, prior-art synthesis, routing/trigger boundaries, trigger or output evals, Skill IR, release gates, README/Profile preparation, GitHub repository and pull-request publication, versioned Releases, clean npx installation, team reuse, and create-and-publish flows. The publication path is self-contained and forbids direct default-branch pushes. Exclude one-off summaries, translations, ordinary docs, non-skill package publishing, and tasks that explicitly should not become a skill. metadata: author: Qiaomu version: "2.12.0" upstream_inspiration: yaojingang/yao-meta-skill; joeseesun/qiaomu-skill-publisher --- # Qiaomu Meta Skill Build reusable Qiaomu skill packages, not long prompts. ## Router Rules - Route by frontmatter `description` first. - Once selected, `qiaomu-meta-skill` is the single authoring authority. Do not also invoke a generic `skill-creator` unless the user explicitly requests comparison or this skill is unavailable. - Built-in prior-art discovery belongs to this skill. Do not install, load, or delegate to a separate discovery skill. - Built-in GitHub publishing belongs to this skill. Do not require or invoke a separate publisher skill after this package is selected. - Keep the package root `SKILL.md` to routing and the minimal workflow. Put judgment in `references/`, deterministic behavior in `scripts/`, regression cases in `evals/`, and evidence in `reports/`. - A package has one discoverable root `SKILL.md`; embedded examples and fixtures use `SKILL.example.md` or `SKILL.fixture.md`. - Do not turn one-off summaries, translations, explanations, or brainstorming into skills. - Match the user's action: create/refactor/package requests may edit; audit/evaluate/diagnose-only requests remain read-only; publish only when explicitly requested. - Default to concise Chinese-first `qiaomu-` names with no more than three preferred hyphen parts. - Add `Copyright (c) 向阳乔木`, X `https://x.com/vista8`, and GitHub `https://github.com/joeseesun/` unless another owner is requested. ## Modes - `Scaffold`: exploratory or personal; minimum useful files. - `Production`: team reuse; README, interface, trigger eval, output contract, and install evidence. - `Library`: shared infrastructure; Production plus Skill IR, portability, trust, and review cadence. - `Governed`: public or high-trust; Library plus permission, rollback, secret, release, and claim gates. Choose proportionally with [Operating Modes](references/operating-modes.md), [Gate Selection](references/gate-selection.md), and [QA Ladder](references/qa-ladder.md). ## Built-In Prior-Art Discovery Before a new skill or substantial redesign: