菜单
Create SKILL.md files. Use when building domain skills, packaging workflows, or codifying conventions.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Write conventional commits for the hyperfrontend monorepo — types, scopes (Nx project names), one-line subjects, atomic commits, versioning-harness boundaries. Use when committing, splitting work into commits, choosing a commit scope/type, or asked about commit format.
Register friction found while building hyperfrontend demos into the showcase findings registry. Use when a demo consuming @hyperfrontend/features surfaces an unclear API, confusing error, missing feature, docs gap, packaging snag, or DX papercut — file the finding before writing the workaround.
Write README.md files for hyperfrontend libraries and sub-modules. Use when creating library documentation, writing sub-module READMEs, documenting API tables, adding quick start guides, or reviewing README structure compliance.
Write ARCHITECTURE.md files for hyperfrontend libraries. Use when creating architecture documentation, documenting design decisions, adding system overview diagrams, or explaining module composition and data flow.
Coding conventions for the hyperfrontend Nx monorepo. Use when refactoring, adding features, fixing bugs, or writing any new code across libs/, apps/, tools/, or plugins/.
Achieve 100% code coverage for custom ESLint rules. Use when reviewing ESLint rules for coverage gaps, adding missing tests, or applying istanbul ignore comments.
基于 SOC 职业分类
| name | skill-creator |
| version | 2.0.0 |
| description | Create SKILL.md files. Use when building domain skills, packaging workflows, or codifying conventions. |
| allowed-tools | ["Read","Write","Edit","Grep","Glob","Terminal","AskQuestions"] |
Maximum information compression. Every word earns its place. What readers infer, omit.
Interview first: domain, triggers, pain points, paths, validation, done-criteria → maps directly to skill sections.
Location: .github/skills/<name>/SKILL.md
YAML Frontmatter name, description ("Use when" triggers)
Title One line
Paths Table: role → glob
Gate "Before X: Check Y"
Patterns Copy-paste blocks (no placeholders)
Checklist Done = all checked
Omit what doesn't apply.
| Principle | ✗ Verbose | ✓ Compressed |
|---|---|---|
| Imperative | "You should use..." | "Use" |
| Show don't tell | "The function should report" | "Reports" |
| Examples > prose | "Format paths as tables because…" | | Rules | src/rules/*.ts | |
| Inference | "Since TypeScript compiles to…" | (omit—reader knows) |
| Signpost | Re-explain AST traversal here | See eslint-rules skill |
| Bold = mandatory | "Consider reusing" | MUST REUSE |
| Code = exact | doThing(<your-value>) | doThing('actual-value') |
Test: Delete a word. Meaning lost? Keep it. Otherwise, cut.
Agent sees only description when choosing. ≤1024 chars · Third person · "Use when [triggers]."
✓ Extract/fill/merge PDFs. Use when working with PDF files or forms.
✗ Helps with documents.
Include: this repo's paths, conventions, gotchas. Exclude: domain fundamentals, language syntax, tool installation—reader knows or finds elsewhere.
Split to REFERENCE.md at ~100 lines or distinct sub-domains.
Add script when: deterministic, repeated, needs explicit error handling.
name = folder name