Skip to content
Back to search

qiaomu-meta-skill

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.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/joeseesun--qiaomu-meta-skill/badge.svg)](/listings/joeseesun--qiaomu-meta-skill)

HTML

<a href="/listings/joeseesun--qiaomu-meta-skill"><img src="/api/listings/joeseesun--qiaomu-meta-skill/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 162 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. ---
  2. name: qiaomu-meta-skill
  3. description: |
  4. 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.
  5. metadata:
  6. author: Qiaomu
  7. version: "2.12.1"
  8. upstream_inspiration: yaojingang/yao-meta-skill; joeseesun/qiaomu-skill-publisher
  9. ---
  10. # Qiaomu Meta Skill
  11. Build reusable Qiaomu skill packages, not long prompts.
  12. ## Router Rules
  13. - Route by frontmatter `description` first.
  14. - 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.
  15. - Built-in prior-art discovery belongs to this skill. Do not install, load, or delegate to a separate discovery skill.
  16. - Built-in GitHub publishing belongs to this skill. Do not require or invoke a separate publisher skill after this package is selected.
  17. - 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/`.
  18. - A package has one discoverable root `SKILL.md`; embedded examples and fixtures use `SKILL.example.md` or `SKILL.fixture.md`.
  19. - Do not turn one-off summaries, translations, explanations, or brainstorming into skills.
  20. - Match the user's action: create/refactor/package requests may edit; audit/evaluate/diagnose-only requests remain read-only; publish only when explicitly requested.
  21. - Default to concise Chinese-first `qiaomu-` names with no more than three preferred hyphen parts.
  22. - Add `Copyright (c) 向阳乔木`, X `https://x.com/vista8`, and GitHub `https://github.com/joeseesun/` unless another owner is requested.
  23. ## Modes
  24. - `Scaffold`: exploratory or personal; minimum useful files.
  25. - `Production`: team reuse; README, interface, trigger eval, output contract, and install evidence.
  26. - `Library`: shared infrastructure; Production plus Skill IR, portability, trust, and review cadence.
  27. - `Governed`: public or high-trust; Library plus permission, rollback, secret, release, and claim gates.
  28. Choose proportionally with [Operating Modes](references/operating-modes.md), [Gate Selection](references/gate-selection.md), and [QA Ladder](references/qa-ladder.md).
  29. ## Built-In Prior-Art Discovery
  30. Before a new skill or substantial redesign:
  31. 1. Derive 2–4 intent-shaped queries covering outcome, domain action, quality mechanism, and an adjacent synonym when useful.
  32. 2. Prefer the unified runner:
  33. ```bash
  34. python3 scripts/research_prior_art.py "<query 1>" "<query 2>" --strict --summary --output reports/prior-art-candidates.json
  35. ```
  36. Its underlying catalog calls remain:
  37. ```bash
  38. npx --yes skills find "<query>"
  39. python3 scripts/search_skillsmp.py "<query>" --limit 20 --sort stars
  40. ```
  41. 3. Keep metrics separate: skills.sh installs measure adoption; SkillsMP stars belong to the source repository; neither is a user rating or quality score.
  42. 4. Deduplicate by canonical GitHub repository and skill path. Collapse translations, mirrors, and obvious forks without adding metrics together.
  43. 5. Shortlist genuinely relevant popularity, trust, and complementary anchors. Inspect source `SKILL.md`, maintenance, license, permissions, security signals, and available rating evidence; never execute untrusted candidate code just to study it.
  44. 6. Synthesize `keep / adapt / reject / invent`. Map each adopted mechanism to the new package instead of collaging prose.
  45. 7. Preserve dated sources, metrics, failures, deduplication, lessons, rejections, and missing evidence in `reports/prior-art-research.md` for Production+ or materially researched work.
  46. If a catalog fails, continue with the other sources, record `missing evidence`, and lower the claim. Full method: [Prior-Art Research](references/prior-art-research.md).
  47. ## Generalization Gate
  48. Before promoting one failure into a core rule:
  49. 1. restate it as a domain-neutral behavior
  50. 2. classify it as core mechanism, optional adapter, or eval-only fixture
  51. 3. promote only safety/factual/permission invariants or behavior repeated across unrelated domains
  52. 4. keep one-off details in fixtures or specialist references
  53. 5. rerun the original and unrelated boundary cases
  54. Prefer intent fidelity, source fidelity, and decision rules over an expanding topic encyclopedia.
  55. ## Qiaomu Skill OS
  56. 1. `Intent`: recurring job, users, inputs, output, exclusions, standards, references.
  57. 2. `Skill IR`: platform-neutral meaning and evidence boundary.
  58. 3. `Package`: lean root instructions, interface, README, and earned resources.
  59. 4. `Eval`: trigger boundaries first; output/runtime/human eval when risk justifies it.
  60. 5. `Review`: package, context, trust, install, README, and public claims.
  61. 6. `Operate`: explicit feedback, failures, drift, and next-iteration proposals without raw private content.
  62. ## Compact Workflow
  63. 1. Decide whether the request deserves a reusable skill; otherwise answer directly and create no package.
  64. 2. Capture job, finished output, target users, inputs, exclusions, permissions, standards, existing assets, platforms, and publication intent.
  65. 3. Pass prior-art discovery or record why it is not applicable or missing evidence.
  66. 4. Pass the generalization gate for sample-driven core changes.
  67. 5. Choose the lightest valid mode.
  68. 6. Write the `description` early (quote it or use `description: |` when it contains `: `, e.g. `Excludes:`); run `evals/trigger_cases.json` before expanding structure.
  69. 7. Create only earned resources. Never create ceremonial directories or duplicate README/SKILL prose.
  70. 8. Export `reports/skill-ir.json` for Production+, public, or cross-platform packages.
  71. 9. Add output evals when correctness, safety, persuasion, or repeatability cannot be shown by trigger tests alone.
  72. 10. Keep mutations within the requested action boundary and preserve rollback for risky changes.
  73. 11. Validate package, unit tests, trigger behavior, context budget, secret/trust boundaries, and evidence claims.
  74. 12. Produce the creation handoff and clearly label missing evidence.
  75. 13. When publication is requested, read [Self-Contained Skill Publishing](references/publishing.md), then use the bundled publisher for feature branch → validation → PR → merge → release/install verification; never push directly to the default branch.
  76. Core commands:
  77. ```bash
  78. python3 scripts/validate_skill.py .
  79. python3 scripts/export_skill_ir.py . --output reports/skill-ir.json
  80. python3 scripts/trigger_eval.py . --cases evals/trigger_cases.json --output reports/trigger-eval.json
  81. python3 scripts/release_check.py . --phase local --run-tests
  82. python3 scripts/publish_skill.py /path/to/skill --dry-run
  83. ```
  84. ## Gate Ladder
  85. - `Scaffold`: valid frontmatter, useful README hook, natural triggers, explicit exclusions.
  86. - `Production`: Scaffold plus interface, trigger eval, output contract, troubleshooting, root isolation, and install verification.
  87. - `Library`: Production plus Skill IR, portability, trust, review cadence, and evidence artifacts.
  88. - `Governed`: Library plus permission/rollback boundary, secret scan, output or integrity-preserving human evidence, and public-claim guard.
  89. Unavailable telemetry, provider runs, approval, install proof, or human review must remain `missing evidence`; planned work is not proof. See [Review And Release Gates](references/review-release-gates.md) and [Resource Boundary Spec](references/resource-boundaries.md).
  90. ## Output Contract
  91. For package-producing requests, provide only what the selected mode earns:
  92. 1. working skill directory and trigger-aware root `SKILL.md`
  93. 2. aligned `agents/interface.yaml`
  94. 3. human-facing README for shared/public skills
  95. 4. trigger cases and generated trigger report for Production+
  96. 5. Skill IR, prior-art report, and creation handoff for Production+
  97. 6. optional references, scripts, output evals, reports, and manifest when they improve judgment, repeatability, or evidence
  98. 7. publish artifacts only when publishing was requested
  99. The final creation handoff must name the **reference skills studied**, give **candidate-specific lessons**, explain deliberate rejections and original contributions, and label each highlight as **design advantage**, **validated advantage**, or **hypothesis**. Never claim global superiority without a fair comparison. Use [Creation Handoff](references/creation-handoff.md).
  100. ## Publish Flow
  101. 1. Treat README as a product page: value, install, natural examples, prerequisites, outputs, configuration, risks, and troubleshooting.
  102. 2. Audit without mutation when useful: `python3 scripts/publish_skill.py /path/to/skill --dry-run`.
  103. 3. Only after an explicit publish request, run `python3 scripts/publish_skill.py /path/to/skill`.
  104. 4. The bundled publisher prepares MIT LICENSE, README and Qiaomu profile assets; resolves skill/repository identity; blocks secrets and reused release versions; creates or reuses a GitHub repository; and publishes only through a feature branch and PR.
  105. 5. Merge is blocked by conflicts, failed/pending checks or requested changes. Successful publication creates `vX.Y.Z`, verifies `npx skills add --list`, performs an isolated install, and runs the published release gate.
  106. 6. Do not report publication complete until the remote default version, GitHub Release, discovery and clean installation are verified.
  107. Detailed CLI and safety decisions: [Self-Contained Skill Publishing](references/publishing.md). README method: [GitHub README Playbook](references/github-readme-playbook.md). Operation method: [SkillOps Loop](references/skillops-loop.md).
  108. ## Qiaomu Defaults
  109. - Prefer practical, concise, publishable Chinese output.
  110. - Keep one creator authority and one root skill entrypoint.
  111. - Preserve platform-neutral source plus minimal adapters.
  112. - Public claims must match trigger, output, runtime, install, or human evidence actually present.
  113. - Anything a skill makes that a person will see (installer pages, dialogs, generated documents) must be rendered and looked at before it is called verified; structure and signature checks do not count.
  114. - When a skill encodes an aesthetic choice (text styles, colours, motion, layouts), prototype at least three restrained directions on the real target content with the real renderer, compare them side by side, then code the winner; never stack effects to look "designed". Run slow or external gates (notarization, review queues) only after a fast local check has passed.
  115. - Upstream ideas are adopted semantically with attribution, not mirrored wholesale; check the license first, and never copy copyleft (GPL) code into a permissively licensed package.
  116. - Machine-specific facts a skill records (keychain profile names, local paths, installed tools) say where they live and how to recreate them; secrets and passwords are never written into a skill.
  117. - Lessons from a real project go into the matching reference of the existing skill, not a new skill.
  118. - Visual-output skills must get diversity from structure, not skin: per-unit visual metaphor, mixed layout topologies, style defined as narrative grammar axes, and a runnable check seeded with a real failed output; see item 20 of [Skill Engineering Method](references/skill-engineering-method.md).
  119. - Long-running, output-producing skills declare size tiers with time estimates, ship graded examples, and gate on checks that must reach zero; see items 17–19 of [Skill Engineering Method](references/skill-engineering-method.md).
  120. ## Reference Map
  121. - Design: [Skill Engineering Method](references/skill-engineering-method.md), [Skill Archetypes](references/skill-archetypes.md), [Intent Dialogue](references/intent-dialogue.md), [Non-Skill Decision Tree](references/non-skill-decision-tree.md)
  122. - Evidence: [Eval Playbook](references/eval-playbook.md), [Output Eval](references/output-eval-method.md), [Skill IR](references/skill-ir-method.md), [Governance](references/governance.md)
  123. - Release: [Self-Contained Publishing](references/publishing.md), [Review And Release Gates](references/review-release-gates.md), [GitHub README](references/github-readme-playbook.md), [SkillOps](references/skillops-loop.md)
README.md · README · 360 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. # qiaomu-meta-skill
  2. > 把一句「把这个流程做成 Skill」,变成一个真正能被发现、能稳定触发、能通过验证、还能一键开源的 Skill。
  3. [![GitHub Release](https://img.shields.io/github/v/release/joeseesun/qiaomu-meta-skill?display_name=tag&sort=semver)](https://github.com/joeseesun/qiaomu-meta-skill/releases)
  4. [![Stars](https://img.shields.io/github/stars/joeseesun/qiaomu-meta-skill?style=flat)](https://github.com/joeseesun/qiaomu-meta-skill/stargazers)
  5. [![Last commit](https://img.shields.io/github/last-commit/joeseesun/qiaomu-meta-skill)](https://github.com/joeseesun/qiaomu-meta-skill/commits/main)
  6. [![License](https://img.shields.io/github/license/joeseesun/qiaomu-meta-skill)](LICENSE)
  7. ```bash
  8. npx skills add joeseesun/qiaomu-meta-skill
  9. ```
  10. 安装以后,你只需要把提示词、SOP、聊天记录、旧 Skill、脚本或一个模糊想法交给 Agent:
  11. ```text
  12. 用乔木元 Skill,把这套工作流做成一个可复用的 Skill;
  13. 先研究同类热门 Skill,完成触发评测和安全检查,然后发布到 GitHub。
  14. ```
  15. 它会自己完成:**需求收敛 → 同类检索 → 取长避短 → Skill 设计 → 触发评测 → 格式校验 → README → API 泄露检查 → PR → Release → npx 安装验证**。
  16. **v2.8.1 本地候选已验证:** 35/35 单元测试、23/23 触发评测、0 个包校验问题。公开发布证据以 [Releases](https://github.com/joeseesun/qiaomu-meta-skill/releases) 为准。
  17. ## 为什么我做了这个
  18. Skill 正在变成 Agent 时代真正可复用的软件单元,但“写一份 `SKILL.md`”离一个好用的 Skill 还很远:
  19. - 描述写得太宽,会到处误触发;写得太窄,又永远叫不出来。
  20. - 把一段长 Prompt 换个文件名,不会自动变成可靠工作流。
  21. - 不研究已有方案,很容易重复造一个更差的轮子。
  22. - 本地能跑,不代表别人能安装,更不代表可以安全发布。
  23. - README、许可证、版本、密钥泄露、PR、Release 和安装证明,经常在最后一步一起失控。
  24. Anthropic 与 OpenAI 的官方 `skill-creator` 奠定了很好的基础。乔木元 Skill 在此之上补齐了我实际做几十个 Skill 时最需要的一段:**先搜索再创造、用证据控制质量,并把成品安全发布给别人使用。**
  25. 初始方法来自搭档姚老师的 [`yaojingang/yao-meta-skill`](https://github.com/yaojingang/yao-meta-skill)。我继续研究并整合 Anthropic、OpenAI 等 Agent Skill 的公开最佳实践,随后加入 skills.sh、SkillsMP、GitHub 验源、乔木式轻量门禁与自包含发布能力。
  26. ## 它比普通 Skill 创建器多做什么
  27. | 能力 | 普通“生成 SKILL.md” | qiaomu-meta-skill |
  28. |---|---:|---:|
  29. | 从 Prompt / SOP / 对话 / 旧 Skill 提炼工作流 | ✓ | ✓ |
  30. | 先搜索 skills.sh 与 SkillsMP 的相关 Skill | | ✓ |
  31. | 回到 GitHub 核对来源、维护、安全与许可证 | | ✓ |
  32. | 记录 `keep / adapt / reject / invent`,避免拼贴抄袭 | | ✓ |
  33. | 测试该触发与不该触发的真实说法 | 视实现而定 | ✓ |
  34. | 区分设计优势、已验证优势和待验证假设 | | ✓ |
  35. | 校验目录、版本、上下文预算与递归发现 | | ✓ |
  36. | README、MIT License、乔木 Profile 自动准备 | | ✓ |
  37. | Secret / API 泄露扫描 | | ✓ |
  38. | 功能分支、PR、检查、Release | | ✓ |
  39. | `npx skills add` 公开发现与隔离安装验证 | | ✓ |
  40. 它不是让 Skill 变得更重,而是让复杂度与风险匹配:个人试验走轻量 `Scaffold`,公开发布才启用完整 `Governed` 门禁。
  41. ## 真实做出来过什么
  42. 截至 2026-08-04,我扫描并去重了本机 Codex 会话、创建交接和 prior-art 报告。能确认有明确创建或实质重构证据的 Qiaomu Skill 共 **28 个**;其中 **18 个已有公开仓库**。这不是“可能适用”的演示列表,而是真实对话留下的工作结果。
  43. ### 已公开,可直接查看
  44. | Skill | 它解决什么问题 |
  45. |---|---|
  46. | [`qiaomu-campus-resume`](https://github.com/joeseesun/qiaomu-campus-resume) | 一问一答深挖大学生经历,生成 ATS 友好的精美 PDF 简历 |
  47. | [`qiaomu-course-designer`](https://github.com/joeseesun/qiaomu-course-designer) | 通过依赖感知访谈,把模糊课程想法收敛成课程蓝图 |
  48. | [`qiaomu-ppt`](https://github.com/joeseesun/qiaomu-ppt) | 从资料研究、大纲到可编辑、可验证的 PPT / HTML Deck |
  49. | [`qiaomu-bento-ppt`](https://github.com/joeseesun/qiaomu-bento-ppt) | 独立生成和编辑 Bento 风格演示文稿 |
  50. | [`qiaomu-cover-designer`](https://github.com/joeseesun/qiaomu-cover-designer) | 从 URL 或内容生成多风格高级概念封面 |
  51. | [`qiaomu-book-script`](https://github.com/joeseesun/qiaomu-book-script) | 把非虚构书籍提炼成能让人停下手指的口播稿 |
  52. | [`qiaomu-drama-generator`](https://github.com/joeseesun/qiaomu-drama-generator) | 生成中文竖屏短剧的人设、大纲与完整剧本 |
  53. | [`qiaomu-xinzhiyuan-title`](https://github.com/joeseesun/qiaomu-xinzhiyuan-title) | 基于真实语料学习新智元风格的 AI 科技标题 |
  54. | [`qiaomu-read-helper`](https://github.com/joeseesun/qiaomu-read-helper) | 用飞书章节、划线和评论完成共读与读书笔记 |
  55. | [`qiaomu-goal-meta-skill`](https://github.com/joeseesun/qiaomu-goal-meta-skill) | 把模糊任务收敛成结果、验证、边界完整的 Codex Goal |
  56. | [`qiaomu-ai-prd`](https://github.com/joeseesun/qiaomu-ai-prd) | 把一句产品想法变成 AI 编程助手可执行的 PRD |
  57. | [`qiaomu-model-cli`](https://github.com/joeseesun/qiaomu-model-cli) | 并发编排 Grok、Kimi 与 Claude Code 等本地模型 CLI |
  58. | [`qiaomu-ai-access`](https://github.com/joeseesun/qiaomu-ai-access) | 检查 AI 服务访问环境信号与合规隐私卫生 |
  59. | [`qiaomu-seo`](https://github.com/joeseesun/qiaomu-seo) | 研究、审计、实施并验证传统搜索与 AI 搜索 SEO |
  60. | [`qiaomu-youtube-download`](https://github.com/joeseesun/qiaomu-youtube-download) | 搜索、下载并验证 YouTube 视频、音频、字幕与元数据 |
  61. | [`qiaomu-wx-video`](https://github.com/joeseesun/qiaomu-wx-video) | 下载并验证微信视频号视频或直播回放 |
  62. | [`qiaomu-music-publisher`](https://github.com/joeseesun/qiaomu-music-publisher) | 从 Suno 下载歌曲、歌词和封面并完成音乐发布工作流 |
  63. | [`qiaomu-meta-skill`](https://github.com/joeseesun/qiaomu-meta-skill) | 元 Skill 自己也持续用同一套研究、评测和发布门禁迭代 |
  64. <details>
  65. <summary><strong>另外 10 个本地或未公开案例</strong></summary>
  66. `qiaomu-vps-website-ops`、`qiaomu-profile`、`qiaomu-xhs-promo`、`qiaomu-xhs-writer`、`qiaomu-kazike-title`、`qiaomu-kazike-writer`、`qiaomu-xinzhiyuan-writer`、`qiaomu-twitter`、`qiaomu-douyin`、`qiaomu-cut`。
  67. 它们只用于证明场景覆盖,不提供不可访问的仓库链接,也不把“本地存在”表述为“已经公开发布”。
  68. </details>
  69. 完整扫描口径与去重清单见 [`reports/codex-skill-catalog.md`](reports/codex-skill-catalog.md)。扫描只输出 Skill 名称、用途和公开状态,不复制私人对话、附件、Token 或本机路径。
  70. ## 它研究过哪些 Skill
  71. 乔木元 Skill 不会看到排行榜第一名就照搬。它会从 skills.sh、SkillsMP 与 GitHub 找出“流行度锚点、可信来源、互补专家”,阅读源文件后再决定保留、改造、拒绝或原创。
  72. <details>
  73. <summary><strong>已进入公开 prior-art 报告的完整去重清单</strong></summary>
  74. ### Skill 创建与评测
  75. - [`anthropics/skills@skill-creator`](https://github.com/anthropics/skills)
  76. - [`openai/skills@skill-creator`](https://github.com/openai/skills)
  77. - [`wshobson/agents@evaluation-methodology`](https://github.com/wshobson/agents)
  78. - [`yaojingang/yao-meta-skill`](https://github.com/yaojingang/yao-meta-skill)
  79. - [`joeseesun/qiaomu-skill-publisher`](https://github.com/joeseesun/qiaomu-skill-publisher)
  80. ### 访谈、课程与简历
  81. - [`alirezarezvani/claude-skills@grill-me`](https://github.com/alirezarezvani/claude-skills/tree/main/engineering/grill-me/skills/grill-me)
  82. - [`mattpocock/skills@grill-me`](https://github.com/mattpocock/skills/tree/main/skills/productivity/grill-me)
  83. - [`addyosmani/agent-skills@interview-me`](https://github.com/addyosmani/agent-skills/tree/main/skills/interview-me)
  84. - [`tyrealq/q-skills@q-educator`](https://github.com/tyrealq/q-skills/tree/main/skills/q-educator)
  85. - [`kevintsai1202/teaching-site-skills@course-outline-design`](https://github.com/kevintsai1202/teaching-site-skills/tree/main/course-outline-design)
  86. - [`pedrohcgs/claude-code-my-workflow@interview-me`](https://github.com/pedrohcgs/claude-code-my-workflow/tree/main/.claude/skills/interview-me)
  87. - [`pedrohcgs/claude-code-my-workflow@syllabus`](https://github.com/pedrohcgs/claude-code-my-workflow/tree/main/.claude/skills/syllabus)
  88. - [`rendercv/rendercv-skill`](https://github.com/rendercv/rendercv-skill)
  89. - [`erichowens/some_claude_skills@cv-creator`](https://skills.sh/erichowens/some_claude_skills/cv-creator)
  90. - [`eachlabs/skills@resume-design-generation`](https://skills.sh/eachlabs/skills/resume-design-generation)
  91. - [`amruthpillai/reactive-resume`](https://github.com/amruthpillai/reactive-resume)
  92. - [`xitanggg/open-resume`](https://github.com/xitanggg/open-resume)
  93. - [`jakegut/resume`](https://github.com/jakegut/resume)
  94. - [`posquit0/Awesome-CV`](https://github.com/posquit0/Awesome-CV)
  95. - [`liantze/AltaCV`](https://github.com/liantze/AltaCV)
  96. - [`tw93/kami`](https://github.com/tw93/kami)
  97. - [`mmmlllnnn/ResumeCollection`](https://github.com/mmmlllnnn/ResumeCollection)
  98. ### 内容、社交与 SEO
  99. - [`autoclaw-cc/xiaohongshu-mcp-skills`](https://github.com/autoclaw-cc/xiaohongshu-mcp-skills)
  100. - [`vivy-yi/xiaohongshu-skills@content-marketing`](https://github.com/vivy-yi/xiaohongshu-skills)
  101. - `vivy-yi/xiaohongshu-skills@copywriting-skills`
  102. - `vivy-yi/xiaohongshu-skills@title-writing`
  103. - `vivy-yi/xiaohongshu-skills@hashtag-optimization`
  104. - [`redfox-data/redfox-community@xiaohongshu-rewrite`](https://github.com/redfox-data/redfox-community)
  105. - [`langchain-ai/deepagents@social-media`](https://github.com/langchain-ai/deepagents)
  106. - [`content-designer/ux-writing-skill`](https://github.com/content-designer/ux-writing-skill)
  107. - [`zc277584121/marketing-skills@content-rewrite`](https://github.com/zc277584121/marketing-skills)
  108. - [`coreyhaines31/marketingskills@seo-audit`](https://github.com/coreyhaines31/marketingskills)
  109. - `coreyhaines31/marketingskills@programmatic-seo`
  110. - `coreyhaines31/marketingskills@ai-seo`
  111. - [`agricidaniel/claude-seo@seo-ecommerce`](https://github.com/agricidaniel/claude-seo)
  112. - [`affaan-m/ECC@seo`](https://github.com/affaan-m/ECC)
  113. - [`firecrawl/firecrawl-workflows@firecrawl-seo-audit`](https://github.com/firecrawl/firecrawl-workflows)
  114. ### 对话中明确要求研究或吸收的项目
  115. - [`hugohe3/ppt-master`](https://github.com/hugohe3/ppt-master)
  116. - [`zarazhangrui/frontend-slides`](https://github.com/zarazhangrui/frontend-slides)
  117. - [`nyblnet/bento`](https://github.com/nyblnet/bento)
  118. - [`yArna/isChinaUser`](https://github.com/yArna/isChinaUser)
  119. - [`larksuite/cli`](https://github.com/larksuite/cli)
  120. - 本地 `baocut`、`gsap`、`lottie` Skill,以及 `qiaomu-mondo-poster-design` 等乔木已有能力
  121. </details>
  122. “研究过”只表示它被纳入有日期的对比与取舍,或在对话中被明确要求查阅,不代表依赖、安装、背书或复制。安装量是采用信号,仓库 stars 是仓库关注度;两者都不是用户评分,也不会被加成一个虚假的总分。
  123. ## 你可以直接这样说
  124. - “把这个提示词升级成一个可以给团队复用的 Skill。”
  125. - “采访我,把这套隐性工作方法整理成 Skill;每次只问一个关键问题。”
  126. - “先搜索同类热门 Skill,分析优缺点,再做一个不抄袭的版本。”
  127. - “优化这个已有 Skill 的触发率、准确性和指令遵循。”
  128. - “审计这个 Skill,只给问题和建议,先不要修改文件。”
  129. - “把这个 Skill 发布到 GitHub,生成 npx 安装命令并验证别人能装。”
  130. ## 它到底会产出什么
  131. 根据场景复杂度,元 Skill 会创建必要而非礼仪性的文件:
  132. ```text
  133. your-skill/
  134. ├── SKILL.md # Agent 路由与最小执行骨架
  135. ├── README.md # 给人看的产品页
  136. ├── LICENSE # 默认 MIT
  137. ├── manifest.json # 版本、作者、平台与门禁
  138. ├── agents/interface.yaml # 跨 Agent 接口
  139. ├── references/ # 长方法、判断与安全边界
  140. ├── scripts/ # 可重复验证与确定性工具
  141. ├── evals/trigger_cases.json # 应触发、不应触发、近邻场景
  142. └── reports/ # Skill IR、研究、评测与发布证据
  143. ```
  144. 个人试验不会被迫拥有整套目录;公开、高风险或团队复用的 Skill 才会逐级增加门禁。
  145. ## 一套完整工作流
  146. 1. **Intent**:确认重复任务、目标用户、输入、输出、边界与成功标准。
  147. 2. **Search**:用 2–4 组意图关键词查询 skills.sh 与 SkillsMP,再回到 GitHub 验源。
  148. 3. **Synthesis**:记录每个候选的 `keep / adapt / reject / invent`,明确原创贡献。
  149. 4. **Package**:写精简 `SKILL.md`,把长判断放进 references,把确定性动作放进 scripts。
  150. 5. **Eval**:先测触发边界;风险需要时再补输出、运行时或人工评测。
  151. 6. **Release**:检查版本、README、许可证、秘密信息与安装入口,经功能分支和 PR 发布。
  152. 7. **Verify**:创建 Release,确认远端默认分支,并在隔离环境完成公开安装。
  153. ## 安装与验证
  154. ```bash
  155. npx skills add joeseesun/qiaomu-meta-skill
  156. ```
  157. 只安装这个 Skill:
  158. ```bash
  159. npx skills add joeseesun/qiaomu-meta-skill --skill qiaomu-meta-skill
  160. ```
  161. 验证:
  162. ```bash
  163. test -f ~/.agents/skills/qiaomu-meta-skill/SKILL.md
  164. python3 ~/.agents/skills/qiaomu-meta-skill/scripts/validate_skill.py \
  165. ~/.agents/skills/qiaomu-meta-skill
  166. ```
  167. 前置条件:
  168. - [ ] Node.js 18+:`node --version`
  169. - [ ] npx 可用:`npx --version`
  170. - [ ] Python 3.9+:`python3 --version`
  171. - [ ] 发布到 GitHub 时安装并登录 GitHub CLI:`gh auth status`
  172. - [ ] 搜索或发布时允许访问 skills.sh、SkillsMP 与 GitHub
  173. ## 内置搜索
  174. ```bash
  175. python3 scripts/research_prior_art.py \
  176. "<query 1>" "<query 2>" \
  177. --strict --summary \
  178. --output reports/prior-art-candidates.json
  179. ```
  180. 底层数据源:
  181. ```bash
  182. npx --yes skills find "<query>"
  183. python3 scripts/search_skillsmp.py "<query>" --limit 20 --sort stars
  184. ```
  185. 详细方法见 [`references/prior-art-research.md`](references/prior-art-research.md)。
  186. ## 自包含发布
  187. 只检查,不改文件、不写 GitHub:
  188. ```bash
  189. python3 scripts/publish_skill.py /path/to/skill --dry-run
  190. ```
  191. 正式发布:
  192. ```bash
  193. python3 scripts/publish_skill.py /path/to/skill
  194. ```
  195. 发布器会依次执行包验证、版本一致性、secret scan、功能分支、PR 检查、合并、GitHub Release、`npx skills add --list`、隔离安装和本地安全同步。
  196. - 不直接推送 `main/master`
  197. - 不覆盖已经发布的同版本 Release
  198. - 不吞掉 push 或检查失败
  199. - 不破坏性删除旧的本地 Skill
  200. - PR 冲突、未完成/失败检查或 requested changes 会阻断自动合并
  201. 完整参数见 [`references/publishing.md`](references/publishing.md)。
  202. ## 本地质量检查
  203. ```bash
  204. python3 scripts/validate_skill.py .
  205. python3 scripts/export_skill_ir.py . --output reports/skill-ir.json
  206. python3 scripts/trigger_eval.py . --cases evals/trigger_cases.json --output reports/trigger-eval.json
  207. python3 scripts/release_check.py . --phase local --run-tests
  208. python3 -m unittest discover -s tests -p 'test_*.py'
  209. ```
  210. ## 常见问题 / Troubleshooting
  211. | 问题 | 常见原因 | 处理方式 |
  212. |---|---|---|
  213. | `No valid skills found` | `SKILL.md` frontmatter 不完整或嵌套入口错误 | 运行 `scripts/validate_skill.py`,修正 `name`、`description` 与根入口 |
  214. | Skill 到处误触发 | description 太泛 | 补 should-not-trigger 与 near-neighbor 用例,收窄描述 |
  215. | Skill 永远不触发 | 用户自然说法没有进入 description | 从真实对话补触发词,再跑 trigger eval |
  216. | README 像内部说明书 | 把 `SKILL.md` 直接复制成 README | 重写成价值、安装、说法、输出、风险与排错 |
  217. | 发布后别人装不上 | 只验证本地目录,没有公开发现和隔离安装 | 完整运行发布器,不把 push 成功当作发布完成 |
  218. | 发布器拒绝版本 | `vX.Y.Z` 已存在 | 提升版本;已发布版本不可覆盖 |
  219. | SkillsMP 网络中断 | 上游分块响应或限流 | 让统一研究器重试并保留 `missing evidence`,不要编造结果 |
  220. ## 设计哲学:Fork 它,而不是膜拜它
  221. Skill 不应该是一套不可修改的“标准答案”。它更像把个人经验编译成 Agent 可以执行的源代码。
  222. 建议先安装、跑一个真实任务,然后 fork:删除不属于你的规则,加入你自己的判断、工具、风格、评测与发布边界。一个越来越像你的 Skill,才真正符合 Skill 的理念。
  223. ## 致谢与来源
  224. - [`yaojingang/yao-meta-skill`](https://github.com/yaojingang/yao-meta-skill):Skill IR、评测证据、Review、信任边界与 SkillOps 方法。
  225. - [`anthropics/skills`](https://github.com/anthropics/skills):Skill 创建、迭代与真实评测实践。
  226. - [`openai/skills`](https://github.com/openai/skills):渐进披露、自由度与可验证的 Skill 打包方法。
  227. - [`joeseesun/qiaomu-skill-publisher`](https://github.com/joeseesun/qiaomu-skill-publisher):README、Profile、License 与安装验证;其能力现已安全内建。
  228. - skills.sh、SkillsMP 与所有在 prior-art 报告中被研究的开源作者。
  229. 上游思想以语义方式吸收并保留归因,不整库镜像,不复制许可证不明的正文,也不把搜索热度冒充质量。
  230. Upstream inspiration: https://github.com/yaojingang/yao-meta-skill; https://github.com/joeseesun/qiaomu-skill-publisher
  231. ## 安全与证据边界
  232. - 公开候选只读取元数据与源码,不会为了学习而执行未经审查的第三方脚本。
  233. - API key、Cookie、Token、私有附件、绝对路径和原始对话不得进入公开仓库。
  234. - 目录安装量、仓库 stars、安全审计和许可证分别记录,不合并为“最佳 Skill 分数”。
  235. - 没有 provider 实跑、人工盲评或用户结果时,必须明确标记 `missing evidence`。
  236. - 发布是外部写操作,只有明确要求时才执行,并通过功能分支、PR、Release 与公开安装验证。
  237. <!-- qiaomu-profile:start -->
  238. ## 关于向阳乔木
  239. 向阳乔木(乔向阳 / Joe)是一位实践型 AI 产品与内容创作者,长期把前沿 AI 变化转译成可复用的工作流、产品判断、AI 编程实践、AI 搜索实践和 GEO/AI 营销方法。
  240. - 个人网站: https://qiaomu.ai
  241. - 博客: https://blog.qiaomu.ai
  242. - X: https://x.com/vista8
  243. - GitHub: https://github.com/joeseesun/
  244. - 微信公众号: 向阳乔木推荐看
  245. ### 支持与关注
  246. | 打赏支持 | 微信公众号 |
  247. |---|---|
  248. | <img src="assets/qiaomu-profile/qiaomu_reward_qr.png" alt="向阳乔木打赏二维码" width="180" /> | <img src="assets/qiaomu-profile/qiaomu_wechat_public_account_qr.jpg" alt="向阳乔木推荐看公众号二维码" width="180" /> |
  249. | 感谢支持乔木持续分享 AI 实践 | 扫码关注「向阳乔木推荐看」 |
  250. <!-- qiaomu-profile:end -->
  251. ---
  252. <a name="english"></a>
  253. ## English
  254. `qiaomu-meta-skill` turns prompts, SOPs, transcripts, scripts, and existing skills into researched, evaluated, installable agent-skill packages.
  255. Unlike a one-shot `SKILL.md` generator, it includes dual-catalog prior-art research, GitHub source verification, trigger evaluation, evidence-aware release gates, secret scanning, pull-request publication, versioned Releases, and clean `npx` installation verification.
  256. ```bash
  257. npx skills add joeseesun/qiaomu-meta-skill
  258. ```
  259. Try saying:
  260. - “Turn this repeated workflow into a reusable skill.”
  261. - “Research the strongest related skills, then synthesize an original version.”
  262. - “Publish this skill to GitHub and prove that a clean machine can discover and install it.”
  263. The project is intentionally fork-friendly: install it, run a real workflow, then replace Qiaomu's defaults with your own judgment, tools, style, and evaluation boundary.
  264. ## License
  265. MIT

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.