Claude-Skills
Pradnyeshp/Claude-Skills
Automação de tarefas de dev: commits, testes, reviews, refatoração
Ver no GitHubColeção de skills para Claude Code que automatizam tarefas comuns de desenvolvimento de software. Inclui geração de mensagens de commit, testes, revisão de código, refatoração, documentação e auditoria de dependências com comando simples.
Instalação
npx convez-ai Pradnyeshp/Claude-Skills1 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
- Gerar mensagens de commit convencionais automáticas
- Criar testes unitários a partir de código modificado
- Revisar diffs para bugs, segurança e estilo
- Debugar e diagnosticar problemas de raiz
- Auditar dependências por vulnerabilidades
Veredito

Suite completa e pragmática de skills que economizam tempo real em workflows de dev. Bem estruturada com 26+ comandos específicos; a qualidade depende da integração com o projeto e do prompt engineering adequado para cada contexto.
README
A tradução para o português deste README ainda está na fila.
Claude-Skills
Collection of helpful Claude Code skills for software development.
Skills
| Skill | Command | Description |
|---|---|---|
| Commit Message | /commit-message | Generate conventional commit messages from staged changes |
| PR Description | /pr-description | Generate PR title and description from branch diff |
| Changelog | /changelog 1.2.0 | Generate changelog entries from recent commits |
| Test Gen | /test-gen | Generate unit tests for changed code, matching the project's test style |
| Review Diff | /review-diff | Review the branch/working-tree diff for bugs, security, and style issues |
| Explain | /explain <file|symbol|diff> | Explain a file, function, or diff in plain language |
| Docstring | /docstring | Add or update doc comments for changed code in the project's style |
| Debug | /debug <symptom> | Diagnose a bug from an error/stack trace/failing test and fix the root cause |
| Refactor | /refactor <target> <goal> | Restructure code toward a goal without changing behavior |
| Scaffold | /scaffold <what> | Generate boilerplate for a new file/component/module from project conventions |
| Deps | /deps | Audit dependencies for outdated, deprecated, or vulnerable packages |
| Onboard | /onboard | Orient to an unfamiliar repo — how to run it, architecture, where to start |
| Regex | /regex <description> | Build, explain, or debug a regex with examples and test cases |
| Optimize | /optimize <target> | Find and fix performance bottlenecks while preserving behavior |
| Tech Debt | /tech-debt | Scan for TODO/FIXME markers and code smells, triaged into an action list |
| Dockerize | /dockerize | Generate a production-ready Dockerfile, .dockerignore, and optional compose |
| CI Setup | /ci-setup | Generate a CI pipeline (GitHub Actions) that lints, tests, and builds |
| SQL | /sql <request> | Write, explain, or optimize a SQL query for the project's dialect |
| Bisect | /bisect <symptom> | Find the commit that introduced a regression via git bisect |
| Migrate | /migrate <change> | Apply a sweeping mechanical change across the codebase, then verify |
| API Docs | /api-docs | Generate OpenAPI or markdown docs from the project's route handlers |
| Gitignore | /gitignore | Generate/tighten .gitignore and flag already-tracked files to untrack |
| Type Annotate | /type-annotate | Add type hints/annotations matching the project's typing strictness |
| Env Example | /env-example | Scan code for env vars and generate/update a safe .env.example |
| Resolve Conflicts | /resolve-conflicts | Walk through and safely resolve git merge/rebase/cherry-pick conflicts |
| Lint Fix | /lint-fix | Detect the project's linter/formatter, run it, and fix the findings |
| Readme | /readme | Generate or update a project README from the codebase |
| Secrets Scan | /secrets-scan | Scan the working tree and git history for hardcoded secrets |
| Release | /release 1.4.0 | Bump version, update changelog, and tag a release |
| Coverage | /coverage | Run tests with coverage and report the riskiest gaps |
| Pre-commit | /pre-commit | Set up pre-commit hooks that lint, format, and check before commits |
| License | /license MIT | Add a LICENSE file (and optional headers) with the right SPDX text |
| Seed | /seed | Generate realistic seed/fixture data matching the project's schema |
Installation
Option 1: Copy to your project (project-level)
Copy the .claude/skills/ folder into your project root:
cp -r .claude/skills/ /path/to/your/project/.claude/skills/
The skills will be available as /commit-message, /pr-description, and /changelog inside that project.
Option 2: Install globally (all projects)
Copy the skill folders to your global Claude config:
cp -r .claude/skills/* ~/.claude/skills/
The skills will be available in every Claude Code session.
Usage
These skills support two invocation methods — use whichever feels natural:
Slash commands (explicit)
/commit-message # generates message from staged changes
/pr-description # generates PR title + body from branch diff
/changelog 2.1.0 # generates changelog entry for version 2.1.0
/test-gen # writes tests for your changed code
/review-diff # reviews the diff for bugs, security, style
/explain src/auth.ts # explains a file, symbol, or "diff" in plain English
/docstring # adds/updates doc comments for changed code
/debug "TypeError at line 42" # diagnoses and fixes a bug from the symptom
/refactor utils.py "split the parser out" # behavior-preserving restructure
/scaffold a UserCard React component # generates boilerplate from conventions
/deps # audits dependencies for outdated/vulnerable packages
/onboard # orients you to an unfamiliar repo
/regex "match an email address" # builds a regex with examples and tests
/optimize parser.js # finds and fixes performance bottlenecks
/tech-debt # scans and triages TODOs and code smells
/dockerize # generates a Dockerfile + .dockerignore for the project
/ci-setup # generates a CI pipeline (GitHub Actions by default)
/sql "top 10 customers by revenue" # writes, explains, or optimizes a SQL query
/bisect "login broke since v1.2.0" # finds the commit that caused a regression
/migrate "replace moment with date-fns" # applies a sweeping change, then verifies
/api-docs # generates API docs from the route handlers
/gitignore # generates/tightens .gitignore for the stack
/type-annotate # adds type hints/annotations to untyped code
/env-example # generates a .env.example from env vars used in code
/resolve-conflicts # walks through and resolves git merge/rebase conflicts
/lint-fix # runs the project's linter/formatter and fixes findings
/readme # generates or updates the project README from the codebase
/secrets-scan # scans the tree and git history for hardcoded secrets
/release 1.4.0 # bumps version, updates changelog, and tags the release
/coverage # runs tests with coverage and reports the riskiest gaps
/pre-commit # sets up pre-commit hooks (lint, format, check)
/license MIT # adds a LICENSE file with the right SPDX text
/seed # generates seed/fixture data matching the schema
Natural language (conversational)
Just ask Claude in plain English — it will automatically detect and invoke the matching skill:
> write a commit message for my staged changes
> what should the commit message be
> prepare a commit
> draft a PR description
> what should the PR say
> open a pull request
> generate release notes
> what changed recently
> prepare a release for version 2.1.0
> write tests for my changes
> what's untested in this file
> review my code before I commit
> spot bugs in the diff
> what does this function do
> walk me through src/auth.ts
> add docstrings to the changed code
> document this function
> why is this test failing
> I'm getting a NullPointerException here
> refactor this function to be more readable
> extract the validation logic into its own module
> scaffold a new API endpoint for health checks
> create a new React component called Sidebar
> are any of my dependencies out of date
> check for vulnerable packages
> help me understand this repo
> how do I run this project
> write a regex for a valid email
> why doesn't this pattern match
> this function is slow, speed it up
> reduce the latency of this endpoint
> find the tech debt in this codebase
> list all the TODOs and FIXMEs
> dockerize this app
> write a Dockerfile for this project
> set up GitHub Actions for this repo
> add a CI pipeline that runs my tests
> write a SQL query for the top 10 customers by revenue
> why is this query slow
> find the commit that broke login
> bisect this regression since the last release
> rename getUser to fetchUser everywhere
> replace every use of moment with date-fns
> document the API endpoints
> generate an OpenAPI spec for these routes
> create a gitignore for this project
> what should I stop tracking in git
> add type hints to this file
> fix the any types in this module
> what env vars does this project need
> generate a .env.example
> help me resolve these merge conflicts
> finish this rebase, I have conflicts
> fix the lint errors
> run the formatter and clean up warnings
> write a README for this project
> my README is out of date, update it
> scan for hardcoded secrets
> did I commit an API key anywhere
> cut a minor release
> bump the version and tag it
> what's my test coverage
> where are the coverage gaps
> set up pre-commit hooks
> run the linter before every commit
> add an MIT license to this project
> what license should I use
> generate seed data for the database
> create fixtures for my models
Both methods produce identical results. Slash commands are faster for muscle memory; natural language is easier when you don't remember the exact command name.
How Claude Code Skills Work
Each skill is a SKILL.md file with YAML frontmatter and markdown instructions. Key features:
- Dynamic context: Lines starting with
!`command`run before Claude sees the skill, injecting live data (e.g.,git diff) - Arguments:
$ARGUMENTScaptures what the user types after the command - Tool permissions:
allowed-toolspre-approves tools so Claude doesn't ask for permission - Auto-invocation:
disable-model-invocation: falseallows Claude to trigger the skill automatically when it matches your natural language prompt. Set totrueif you want slash-command-only invocation.
See Claude Code Skills docs for the full reference.