AI Extensions — Coleção de Skills para Claude Code
dmorand17/ai-extensions
Skills prontos para automação de dev, docs e segurança em Claude Code
Ver no GitHubRepositório pessoal com skills, agents e hooks para Claude Code e agentes Kiro, cobrindo desde controle de versão e release até diagramas, avaliações de segurança e otimização de comunicação. Inclui ferramentas para git-flow, geração de releases, criação de diagramas Excalidraw, entrevistas estruturadas e auditoria de arquitetura.
Instalação
npx convez-ai dmorand17/ai-extensions1 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
- Automatizar fluxo git com conventional commits e PR
- Gerar releases e changelogs semânticos automáticos
- Conduzir entrevistas técnicas estruturadas sobre planos/design
- Auditar segurança e arquitetura de sistemas
- Criar e manter memória de projeto estruturada
Veredito

Coleção madura e bem-estruturada de skills em shell para automação de ciclo de vida de software. Cobre casos reais (git-flow, release, auditorias) com implementação técnica sólida, embora exija familiaridade com Claude Code e shell scripting.
README
A tradução para o português deste README ainda está na fila.
ai-extensions
Personal collection of Claude Code skills, agents, and hooks.
Skills
| Skill | Description |
|---|---|
| git-flow | Conventional Commits + full ship flow on request (branch, commit, push, open PR) |
| release-generator | Cut a GitHub release — suggest the next semver tag, generate categorized release notes, and publish via gh release create (or just generate changelogs/notes) |
| excalidraw-diagram | Create and edit Excalidraw diagrams via MCP |
| markdown-convert | Convert markdown files to other formats using pandoc |
| skill-creator | Create, improve, and evaluate new skills |
| handoff | Compact the current conversation into a handoff document for another agent to pick up |
| caveman | Ultra-compressed communication mode — cuts token usage ~75% by dropping filler while keeping full technical accuracy |
| grill-me | Interview the user relentlessly about a plan or design until reaching shared understanding |
| grill-me-with-docs | Grilling session that challenges plans against the domain model, sharpens terminology, and updates CONTEXT.md/ADRs inline |
| project-memory | Set up and maintain structured project memory in docs/project_notes/ |
| satv2-assessment | Deploy and run AWS SATv2 security assessments using Prowler-based scanning |
| image-enhancer | Enhance image quality — sharpen, upscale, denoise, and clean up compression artifacts |
Agents
| Agent | Description |
|---|---|
| architecture-reviewer | Reviews system architecture and design decisions |
| code-reviewer | Reviews code for quality, correctness, and style |
| docs-explorer | Explores and explains documentation |
| security-auditor | Audits code and infrastructure for security issues |
Hooks
| Hook | Description |
|---|---|
| hook-logger.sh | Logs hook events for debugging extension authoring |
Transparent payload logger for Claude Code hooks. Reads the JSON payload from stdin, appends a timestamped entry to a log file, and always exits 0 so it never blocks the hook chain. Useful as a debugging sidecar while building or troubleshooting hooks.
Setup
Add to your settings.json (or .claude/settings.json):
{
"hooks": {
"PreToolUse": [
{
"matcher": "",
"hooks": [
{ "type": "command", "command": "$HOME/.claude/hooks/hook-logger.sh" }
]
}
]
}
}
Replace PreToolUse with any hook event (SessionStart, PostToolUse, Stop, UserPromptSubmit, etc.). Narrow by tool with "matcher": "Bash" or "matcher": "Edit|Write".
Options
| Flag | Default | Description |
|---|---|---|
-f, --log-file <path> | ~/.claude/logs/hook-debug.log | Log file path |
HOOK_LOGGER_FILE env var | (same) | Override default path (lower precedence than -f) |
Smoke test
echo '{"hook_event_name":"PreToolUse","tool_name":"Bash"}' | ./hooks/hook-logger.sh
tail -n 20 ~/.claude/logs/hook-debug.log
Requires jq for pretty-printed output (falls back to raw JSON if missing).
Installation
git clone git@github.com:dmorand17/ai-extensions.git
cd ai-extensions
Symlink an individual skill into your global skills dir:
ln -s $(pwd)/skills/git-flow ~/.claude/skills/git-flow # or ~/.kiro/skills/...
Symlink everything:
mkdir ~/.claude/{agents,skills,hooks}
for skill in skills/*/; do
ln -sf "$(pwd)/$skill" ~/.claude/skills/$(basename "$skill")
done
for agent in agents/*.md; do
ln -sf "$(pwd)/$agent" ~/.claude/agents/$(basename "$agent")
done
for hook in hooks/*/; do
ln -sf "$(pwd)/$hook" ~/.claude/hooks/$(basename "$hook")
done
Project-scoped install:
ln -s $(pwd)/skills/skill-creator /path/to/project/.claude/skills/skill-creator
Updating
cd ai-extensions && git pull
Symlinks pick up updates automatically.
Structure
skills/ # Self-contained skill packages, each with a SKILL.md
agents/ # Subagent definitions (one .md file each)
hooks/ # Hook scripts
Each skill directory contains SKILL.md plus any templates, examples, or references it needs.
Useful external skills
Skills from other authors that pair well with this repo:
| Name | Author | Description |
|---|---|---|
| caveman | JuliusBrussee | Ultra-compressed communication mode plus a caveman-compress skill that compresses memory files (CLAUDE.md, todos) to save input tokens |
Inspiration
Projects and writing that shaped how skills here are designed:
| Source | Why |
|---|---|
| mattpocock/skills | Productivity and engineering skills — caveman, grill-me, handoff, grill-with-docs |
Credits
- satv2-assessment — based on Running AWS SATv2 Security Assessments with Kiro Skills