claude-code-skill-factory — Установка & Живое демо

Зачем использовать

Ключевые функции

5 interactive builders covering every Claude Code extension shape
/build slash command — end-to-end wizard for skills, agents, prompts, hooks
/validate-output — checks SKILL.md frontmatter, file layout, packages as ZIP
9 production-ready skills included out of the box
/sync-agents-md generates AGENTS.md for cross-tool compatibility (Cursor, Codex)
Templates based on 69+ preset prompt frameworks

Живое демо

Как выглядит на практике

claude-code-skill-factory-skill.replay ▶ готово

0/0

Установка

Репозиторий автора · Официальная документация Claude Skills

Сценарии использования

Реальные сценарии: claude-code-skill-factory

How to build and package a new Claude Skill in 10 minutes

👤 Developers who want to ship a skill without hand-writing SKILL.md boilerplate ⏱ ~15 min beginner

Когда использовать: You have a repeatable workflow in your head and want Claude to package it as an installable skill.

Поток

Invoke /build

/build — I want a skill that audits a Next.js app for Core Web Vitals issues and outputs a Lighthouse-style report.✓ Скопировано

→ Factory asks clarifying questions: triggers, tool categories, output format
Answer the scaffolding questions

Trigger on 'audit CWV' or Next.js project detected. Output markdown report with screenshots.✓ Скопировано

→ Draft SKILL.md with frontmatter, example usages, reference files
Validate and package

/validate-output✓ Скопировано

→ Green checks on frontmatter, directory structure, example completeness; ZIP produced

Итог: A working, installed skill ready to trigger on the next matching prompt.

Подводные камни

Skill triggers too aggressively on unrelated prompts — Tighten the 'description' in frontmatter — Claude uses it to decide when to load the skill
Example in SKILL.md drifts from actual behavior — /validate-output re-runs the example and flags mismatches

Сочетать с: skills-skill-2

Wire a pre-commit hook via the Hooks Guide

👤 Teams that want to automate Claude Code behavior on specific events ⏱ ~10 min intermediate

Когда использовать: You want something to run automatically on prompt submit, tool use, or session stop — not a skill Claude has to choose to load.

Поток

Start Hooks Guide

/build hook — run gitleaks on every Write tool call before Claude saves secrets✓ Скопировано

→ Guide asks which event: PreToolUse/PostToolUse/Stop
Define matcher and command

Match Write tool calls, pipe file content through gitleaks detect, block on finding.✓ Скопировано

→ Draft settings.json hook block with matcher + command
Install into user settings

/install-hook — user scope✓ Скопировано

→ ~/.claude/settings.json updated; next tool call is gated

Итог: A hook that blocks secret-leaking writes before they happen.

Подводные камни

Hook runs too slow and stalls every tool call — Move to PostToolUse if the check doesn't need to block

Сочетать с: claude-code-owasp-skill

Author a reusable prompt pack using research-backed frameworks

👤 Prompt engineers building consistent house prompts for a team ⏱ ~15 min beginner

Когда использовать: You want every teammate to get CO-STAR / RISEN-structured prompts without memorizing the frameworks.

Поток

Invoke Prompts Guide

/build prompts — 5 presets for customer-support replies, CO-STAR structured✓ Скопировано

→ Guide lists available frameworks and asks which tone/role bias

Итог: Team-shared prompt pack installable via one git clone.

Сочетать с: prompt-architect-skill

Комбинации

Сочетайте с другими MCP — эффект x10

claude-code-skill-factory-skill + prompt-architect-skill

Prompt Architect tunes individual prompts; Skill Factory packages sets of them into a shareable skill

Have prompt-architect tighten these 5 prompts, then /build prompts and export as a skill pack.✓ Скопировано

claude-code-skill-factory-skill + skills-skill-2

Use Posit's skill collection as reference templates when building your own

Show me the structure of the Posit R-analysis skill, then /build a parallel Python-analysis skill modeled on it.✓ Скопировано

Инструменты

Что предоставляет этот MCP

Инструмент	Входные данные	Когда вызывать	Стоимость
/build	target: 'skill'\|'agent'\|'prompt'\|'hook', description: str	Top-level entry to create any new extension	interactive Claude session
/validate-output	path: str	Before distributing a skill	0
/install-skill	zip_or_path: str, scope?: 'user'\|'project'	After validate passes	0
/install-hook	hook_definition	After Hooks Guide produces a draft	0
/sync-agents-md	skill_path: str	Distributing across Cursor/Codex	0
/factory-status	-	Troubleshooting missing builders	0

Стоимость и лимиты

Во что обходится

Квота API: None — runs entirely inside Claude Code
Токенов на вызов: 5k-20k tokens for a full build session
Деньги: Free. Consumes your Claude subscription.
Совет: Keep builder sessions focused — answer decisively to avoid expanding the conversation into rework.

Безопасность

Права, секреты, радиус поражения

Хранение учётных данных: No credentials. Skill Factory writes to ~/.claude/skills and ~/.claude/settings.json only.

Исходящий трафик: No outbound calls beyond Claude Code itself.

/install-hook modifies settings.json — review the generated hook before installing to user scope.

Устранение неполадок

Частые ошибки и исправления

Generated ZIP fails /install-skill

Usually a frontmatter typo in SKILL.md. Run /validate-output first; it surfaces the exact missing field.

Hook produces no output after install

Matcher is too narrow. /factory-status shows which events are firing; widen the matcher in settings.json.

Проверить: cat ~/.claude/settings.json | jq .hooks

Альтернативы

claude-code-skill-factory в сравнении

Альтернатива	Когда использовать	Компромисс
skills-skill-2 (Posit skills)	You just want ready-made skills, not a factory to build your own	No scaffolding, no hook builder
Hand-writing SKILL.md	You've already built several skills and know the shape	More control, slower, no validation

Ещё

Ресурсы

📖 Читать официальный README на GitHub

🐙 Открытые задачи

🔍 Все 400+ MCP-серверов и Skills