Skill CEFailures
brocef/skill-cefailures
Skills de referência de bibliotecas com gerador automático a partir de docs
Ver no GitHubColeção de skills Claude para documentação e troubleshooting de bibliotecas específicas. Fornece conhecimento de APIs, padrões de uso e guias de debug para agentes de código. Funciona em Claude Code, Codex e Desktop app.
Instalação
npx convez-ai brocef/skill-cefailures1 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
- Auxiliar agentes de IA em debugging e troubleshooting de bibliotecas
- Consultar padrões de uso e APIs de forma estruturada
- Gerar automaticamente skills a partir de documentação online
Veredito

Ferramenta técnica sólida para estruturar conhecimento de bibliotecas em skills reutilizáveis. Suporta múltiplos backends (OpenAI, Anthropic, Claude CLI) e automatiza criação via documentação. Útil para desenvolvedores que querem enriquecer agentes Claude com contexto específico de bibliotecas.
README
A tradução para o português deste README ainda está na fila.
skill-cefailures
Agent skills for specific libraries, plus tooling to create new skills from online documentation. The skills can be used from Claude Code and Codex.
Each library gets a skill that provides API/pattern knowledge and debugging/troubleshooting guidance to an AI coding agent.
Installation
As a plugin (recommended)
In Claude Code, run:
/plugin marketplace add brocef/skill-cefailures
/plugin install skill-cefailures
Skills are available as /skill-cefailures:<skill-name> (e.g. /skill-cefailures:brain-style).
On claude.ai/code or in the Claude Desktop app, open the plugin directory, choose Add marketplace, and enter brocef/skill-cefailures.
In Codex, add the marketplace and install the plugin:
codex plugin marketplace add brocef/skill-cefailures --ref main
codex plugin add skill-cefailures@skill-cefailures
Skills are available as skill-cefailures:<skill-name> in Codex sessions after starting a new thread.
Local development
If you've cloned the repo, you can load it directly:
claude --plugin-dir /path/to/skill-cefailures
Creating a new skill
pip install -r requirements.txt
# Using OpenAI (default — requires OPENAI_API_KEY)
python scripts/create_skill.py --name redis --url "https://example.com/redis-docs.md"
# Using Anthropic SDK (requires ANTHROPIC_API_KEY)
python scripts/create_skill.py --name redis --url "https://example.com/redis-docs.md" --backend sdk
# Using the local Claude CLI (no API key needed; requires Claude Code installed)
python scripts/create_skill.py --name redis --url "https://example.com/redis-docs.md" --backend cli
This fetches the documentation, uses an LLM to analyze and split it into a SKILL.md routing layer plus topical reference docs, and writes everything to skills/<name>/.
Backend choices:
--backend openai(default) — uses the OpenAI API. Requirespip install openaiandOPENAI_API_KEY(or--api-key).--backend sdk— uses the Anthropic API directly. Requirespip install anthropicandANTHROPIC_API_KEY.--backend cli— shells out to the localclaudeCLI (requires Claude Code).
Optional flags: --model <id> overrides the per-backend default model; --force overwrites an existing skill directory.
Repo Structure
.agents/plugins/
marketplace.json # Codex marketplace listing
.claude-plugin/
plugin.json # Plugin manifest (name, version, skills/commands paths)
marketplace.json # Marketplace listing
.codex-plugin/
plugin.json # Codex plugin manifest
commands/
brain-style/ # /skill-cefailures:brain-style:* commands
permissions-auditor/ # /skill-cefailures:permissions-auditor:* commands
skills/ # One directory per skill
brain-style/
capabilities-sdlc/
report/
<name>/
SKILL.md # Routing layer (loaded on invocation)
docs/
<topic>.md # Detailed reference (read on demand)
scripts/
create_skill.py # Generate skill from URL
analyze_permissions.py # Analyze permission request logs
apply_permissions.py # Apply curated permission lists to settings.json
log-permission-requests.sh # Permission logging hook script
tests/ # pytest suite
docs/
release-notes/ # Per-version user-facing notes (vX.Y.Z.md + upcoming.md)
changelogs/ # Per-version developer changelogs
plans/ # Design and implementation documents
AGENTS.md # Project instructions for Codex and compatible agents
CLAUDE.md # Symlink to AGENTS.md for Claude Code compatibility
requirements.txt # Runtime dependencies
requirements-dev.txt # Runtime + test dependencies (use this to run the suite)
Skills
| Skill | Purpose |
|---|---|
brain-style | Code style preferences across TypeScript naming, types, and resource-oriented HTTP route design. |
capabilities-sdlc | Project-neutral user-capability documentation conventions for co-located or centralized files, with a planning gate, contradiction-detection rule, and optional multi-repo coordination. |
report | How to report a bug, issue, or suggestion about this plugin — files a GitHub issue on this repo, with a ready-to-fill skeleton for each kind of report. |
Slash Commands
This plugin ships slash commands under commands/ (registered via plugin.json). Commands are for workflows that are purely user-explicit — they replace the auto-trigger metadata of skills, which keeps skill descriptions narrow and saves session context.
| Command | Purpose |
|---|---|
/skill-cefailures:brain-style:review | Architecture review of a file/function/class against decomposition, file-size, and redundancy guidelines. |
/skill-cefailures:brain-style:agents-md | Review or author a project's AGENTS.md against the minimal-routing principle and required-sections checklist, and ensure CLAUDE.md is a symlink to it. |
/skill-cefailures:permissions-auditor:install | Install the permission-logging hook so future permission prompts are captured for later triage. |
/skill-cefailures:permissions-auditor:analyze | Analyze logged permission requests, present recurring patterns, and triage them into allow/deny/manual-review rules. |
How Skills Work
When an agent invokes a library skill:
- SKILL.md is loaded — gives the agent a brief overview and a routing table to reference docs in
docs/. The frontmatterdescriptionfield acts as the skill's invocation criteria. - The agent reads specific docs/ files based on the current task — only what's needed
- Reference docs contain full API details, examples, and troubleshooting
License
Apache-2.0. See LICENSE.