Skills do Blog Claude
uygnoey/skills-from-claude-blog
Artefatos extraídos do blog Claude em specs prontas para uso
Ver no GitHubRepositório que converte automaticamente posts oficiais do blog Claude em especificações reutilizáveis (Skills, Agents, Hooks, Guides e Output Styles) seguindo os padrões do Claude Code. Atualizado a cada 4 horas, organiza artefatos por data e slug do post, com suporte multilíngue (EN, KO, ES, JA).
Instalação
npx convez-ai uygnoey/skills-from-claude-blog1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Descobrir padrões e frameworks do Claude documentados oficialmente
- Reutilizar Skills, Agents e Hooks testados da comunidade Anthropic
- Acessar guias multilíngues de deployment e arquitetura
Veredito

Projeto útil de curadoria que democratiza artefatos oficiais do blog Claude em formatos imediatamente reutilizáveis. Excelente como referência e ponto de partida, mas requer familiaridade com specs do Claude Code (SKILL.md, agents, hooks) para máxima aproveitação.
README
A tradução para o português deste README ainda está na fila.
skills-from-claude-blog
Artifacts distilled from posts on claude.com/blog, organized by Claude Code's official extension specs. Not affiliated with Anthropic — this is a third-party summary project.
A repository that converts official Claude blog posts into Claude Code's official specs matching each post's character. A batch runs every 4 hours to reflect new posts and unprocessed ones.
Layout (per blog post)
One blog post = one <YYYY.MM.DD>_<blog-slug>/ folder at the repo root, where the date prefix is the post's published_date and <blog-slug> is the last URL segment. Folders sort naturally from newest to oldest on GitHub. The subfolders below appear conditionally based on the post's character.
<YYYY.MM.DD>_<blog-slug>/
├── description.en.md # Human-readable English summary (always)
├── description.ko.md # Korean summary (always)
├── description.es.md # Spanish summary (always)
├── description.ja.md # Japanese summary (always)
├── source.json # Source metadata (always)
│
├── skills/<name>/SKILL.md # A. Agent Skills spec
├── agents/<name>.md # B. Claude Code Subagent spec
├── guides/<name>.{en,ko,es,ja}.md # C. Free-form multilingual guides
├── hooks/<name>.json +.md # D. Claude Code Hooks JSON + notes
├── output-styles/<name>.md # E. Output Style spec
└── plugin/ # G. Plugin bundle (rare)
├── .claude-plugin/plugin.json
└── skills|agents|hooks|output-styles/...
Official spec references
- Agent Skills —
SKILL.md - Claude Code Subagents —
agents/<name>.md - Claude Code Hooks —
hooks/<name>.json - Output Styles —
output-styles/<name>.md - Plugins —
plugin/bundle
Spec selection matrix
| Verdict | Trigger question | Artifact |
|---|---|---|
| A. Skill | Does the post describe a reusable pattern / principle / framework / how-to? | skills/<name>/SKILL.md |
| B. Subagent | Does the post explicitly define 2+ named agent roles? | agents/<name>.md each |
| C. Guide | Is the post deployment / architecture / methodology / survey in character? | guides/<name>.{en,ko,es,ja}.md |
| D. Hook | Does the post automate a lifecycle event (PreToolUse, PostToolUse, Stop, SessionStart…)? | hooks/<name>.json + .md notes |
| E. Output Style | Does the post define a full system-prompt tone/role/format swap (not a one-line tip)? | output-styles/<name>.md |
| G. Plugin | Does the post center on bundling multiple artifacts for distribution? | whole plugin/ bundle |
- F. Slash Commands are not produced. Official docs mark
.claude/commands/as legacy and recommend.claude/skills/, so posts covering slash commands are converted into Skills. - Multiple verdicts applying at once is normal. Produce all.
- When uncertain, default to including A. Skills are the most practically reusable.
- Never invent roles, patterns, or scripts that aren't in the source. If unsure, quote the source.
Drop-in usage
Each artifact can be copied into a project as-is.
| Artifact | Destination |
|---|---|
skills/<name>/ | .claude/skills/<name>/ or ~/.claude/skills/<name>/ |
agents/<name>.md | .claude/agents/<name>.md or ~/.claude/agents/<name>.md |
hooks/<name>.json content | Merge into the hooks field of .claude/settings.json |
output-styles/<name>.md | .claude/output-styles/<name>.md |
plugin/ | Load with --plugin-dir ./plugin |
Index
All guides and summaries are available in English, Korean, Spanish, and Japanese. Click the language switcher at the top of each file.
Batch operation
- A Perplexity Computer cron runs every 4 hours.
- Post list is harvested from
https://www.claude.com/sitemap.xmland the/blogindex. - Priority: new since last run (newest → oldest) → unprocessed older posts (oldest → newest) → unknown pub-date.
- Up to 2 posts per run.
Authoring guidelines
SKILL.mdandagents/*.mdfollow the official spec (English).- Hook JSON reflects the post's behavior verbatim; shell commands cite the source in comments.
- Output Styles copy the post's tone/role/format; if unsure, downgrade to a guide.
namefields (Skill, Subagent, Output Style, Plugin):^[a-z0-9-]+$, ≤64 chars, forbidden reserved wordsclaude,anthropic.descriptionis third-person and ≤1024 chars (Skill / Subagent).guides/are authored in all four languages (.en.md,.ko.md,.es.md,.ja.md) with a language switcher at the top.- Human summaries (
description.*.md) cover the same four languages with a language switcher. - Never invent content not in the source. If unsure, fall back to "see source".
- Artifacts are self-contained.
SKILL.md,agents/*.md,hooks/*.md, andoutput-styles/*.mdmust not reference anything outside their own folder — no../paths, no cross-artifact links, no links back todescription.*.mdorguides/. If outside material is needed, copy it into a local companion file (references/,examples/,templates/,scripts/,prompts/,assets/, or a sibling.md).
Files
.
├── <YYYY.MM.DD>_<blog-slug>/ # One folder per blog post, date-prefixed
├── scripts/
│ ├── list_pending.py # List pending URLs
│ ├── mark_processed.py # Mark a URL processed
│ ├── update_last_run.py # Stamp batch timestamp
│ ├── validate.py # Pre-commit validator for all specs
│ └── strip_parent_refs.py # Rewrite artifacts to drop ../ escapes
└── state/
└── processed.json # Processed URLs + last_run_at
License
- Source posts (Claude blog) are © Anthropic. This repo contains summaries and quotations for study and reference only.
- Repository code: MIT License.