joris887
joris887/exosuit
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/joris887/exosuit --skill skill-create命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | skill-create |
| version | 2.4.0 |
| description | Analyze the repository's technology stack and generate appropriate skills, rules, and hook configurations. |
| trigger | manual |
| depends-on | [] |
| references | [] |
| disable-model-invocation | true |
| user-invocable | true |
| allowed-tools | Read, Glob, Grep, Bash, Edit, Write |
Analyzing repository and generating technology skills.
Identify technologies, frameworks, and tools in use:
Scan for all dependency/config files in the project:
package.json, pyproject.toml, Cargo.toml, Package.swiftgo.mod, pom.xml, build.gradle, Gemfile, composer.json.pre-commit-config.yaml, CI configuration filesIdentify major technology areas by scanning source code imports and configuration files.
Read .claude/skills/SKILLS_INVENTORY.md to avoid creating duplicates.
Categorize each technology by impact and relevance:
| Category | Examples | Skill Priority |
|---|---|---|
| Core Framework (>20% of codebase) | React, Django, Rails, SwiftUI | High — create skill + reference doc |
| Major Library (significant usage) | SQLite, Redis, Prisma | Medium — create skill |
| Build/Dev Tool (development workflow) | webpack, jest, ruff | Low — create skill if complex |
| Minor Dependency (small usage) | lodash, httpx, pydantic | Skip — standard usage, no skill needed |
Create reference docs co-located with the skill (.claude/skills/<tech-name>/references/) when:
Co-locating references with skills keeps everything self-contained and allows relative path references from SKILL.md.
Skip reference doc when:
For each technology that warrants a skill, scaffold using:
bash scripts/init-skill.sh <tech-name>
Execute directly — do NOT read script source first.
.claude/skills/<tech-name>/
├── SKILL.md # Lean (<100 lines): purpose, version, key patterns, pointers
└── references/ # Loaded on demand
├── api.md # API patterns and quick reference
├── gotchas.md # Version-specific issues and workarounds
└── examples.md # Code snippets from the actual codebase
Each generated skill SKILL.md should include:
references/api.md for detailed API patterns" etc.Move detailed content to references/ to keep SKILL.md lean. The context window is a shared resource — only load detail when needed.
# <Technology> Reference (v<version>)
## Usage in This Project
<What we use it for, how it fits in the architecture>
## Key Patterns
<Project-specific patterns with code examples>
## API Quick Reference
<The subset of the API we actually use>
## Configuration
<Our configuration settings and why>
## Gotchas
<Version-specific issues, known bugs, workarounds>
## Official Documentation
<Links to official docs for our version>
For detected file types that don't already have rules:
*.component.tsx, *.service.py), add path-scoped rules for those patternsSave rules to .claude/rules/<rule-name>.md with YAML frontmatter containing paths:.
Based on detected tools:
post-edit-format.sh for the detected formatter.claude/hooks/rules/quality.yaml.claude/hooks/rules/quality.yaml.claude/hooks/rules/quality.yamlUpdate .claude/hooks/ scripts and .claude/settings.json as needed.
For each technology with a reference doc:
After creating all skills:
bash scripts/update-registry.sh to regenerate skills-registry.json.claude/skills/SKILLS_INVENTORY.md with new technology skillsPresent a summary:
### Skill Creation Complete
**Technologies analyzed:** [count]
**Skills created:** [count]
**Reference docs created:** [count]
**Rules created:** [count]
**Hooks configured:** [list]
**Skills skipped (already exists or not needed):** [count]
#### Created Skills:
| Technology | Skill | Reference Doc | Version |
|---|---|---|---|
| [name] | `/[skill-name]` | Yes/No | [version] |
#### Rules Created:
| Rule | Paths | Purpose |
|---|---|---|
| [name] | [patterns] | [what it enforces] |
#### Skipped (standard/minor):
- [list of technologies that didn't warrant skills]
| Bad Output | Why It's Wrong | What To Do Instead |
|---|---|---|
| Generic content Claude already knows | Wastes context window on every invocation | Only include project-specific patterns and gotchas |
| Skill >150 lines without references/ | Exceeds context budget | Split into lean SKILL.md + references/ |
| Examples from training data, not codebase | Doesn't match project conventions | Use actual code from the repo as examples |
| Documenting standard API usage | Claude knows standard APIs | Document project-specific patterns and version gotchas |
.claude/skills/<name>/references/ (co-located with the skill).claude/skills/SKILL_TEMPLATE.md