convez.ai
SkillAutomaçãoAvançado
Sebstrdigital

Claude Tools - Agentes e Skills para Desenvolvimento

Sebstrdigital/claude-tools

Agentes e skills modulares para automação de tarefas técnicas com Claude

Ver no GitHub
InicialPluginScore · de 0 a 100

Coleção de agentes agentic customizados e skills reutilizáveis para Claude, cobrindo auditorias, reviews de código, design de harness, orquestração de trabalho e produtividade. Inclui agentes especializados (scout, builder, heavy, skeptic, grunt) e skills modulares em engenharia, produtividade e utilitários, instaláveis via script automatizado.

Instalação

$npx convez-ai Sebstrdigital/claude-tools

1 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

Estrelas
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 1 semana
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Auditorias de segurança, integridade de dados e arquitetura agentic
  • Reviews profundos de código com análise de padrões e anti-padrões
  • Design e implementação de harness para sistemas autônomos escaláveis
  • Orquestração de trabalho técnico delegando entre agentes especializados
  • Investigação de codebase, debugging e refactoring com reconhecimento

Veredito

Repositório maduro e bem-estruturado de ferramentas enterprise para quem constrói sistemas agentic complexos. Cobertura robusta em auditorias, reviews e padrões, mas requer familiaridade com arquitetura de agentes e CLI; não é para inicialização rápida.

README

A tradução para o português deste README ainda está na fila.

Agentic Tools

Custom agentic AI agents, Claude Code skills, and reference docs by @Sebstrdigital.

What's Included

Agents

Active agents live in agents/ and are installed by ./install.sh.

AgentDescription
scoutRead-only reconnaissance worker for codebase investigation, execution tracing, external repo/doc surveys, and evidence-backed findings.
builderStandard implementation worker for scoped stories, fixes, tests, and refactors with clear validation commands.
heavySenior implementation worker for hard debugging, ambiguous root-cause hunts, cross-cutting refactors, and security-sensitive work.
skepticAdversarial read-only reviewer that tries to refute diffs, claims, and validations before work is accepted.
gruntMechanical-work worker for low-judgment tasks such as renames, doc syncs, formatting sweeps, bulk searches, and known-file installs.

Retired repo-managed agents are removed from installed Claude Code agents by install.sh when they contain this repo's source_id.

Skills

Skills live in category-first folders under skills/ and are installed by ./install.sh to ~/.claude/skills/<skill-name>/. This follows the Matt Pocock-style layout:

skills/
├── engineering/
├── productivity/
└── misc/

Source paths are:

  • skills/engineering/<skill-name>/
  • skills/productivity/<skill-name>/
  • skills/misc/<skill-name>/

Each skill has a required SKILL.md. Supporting files live beside it.

Engineering

SkillDescription
agent-auditQuick pass/fail pre-deployment audit for agentic systems.
agent-reviewDeep agentic architecture review with pattern scoring, anti-pattern scanning, and recommendations.
code-reviewGeneral code quality review across structure, naming, complexity, error handling, and type safety/testing.
data-integrity-auditAudit AI-generated query results from user question through displayed answer.
deep-reviewFour-pass code review chain covering framework correctness, resource safety, security, and adversarial absence detection.
harness-interviewGuided interview capturing domain, systems, platform, autonomy, scale, and harness-type signals.
harness-architectSelects harness type, designs agent decomposition, orchestration, and tool map.
harness-specGenerates an implementation-ready specification for a harness.
harness-buildGenerates runnable scaffolding adapted to harness type and platform.
harness-reviewDeep review of harness implementations against harness patterns and anti-patterns.
infra-auditInterview-based infrastructure assessment with gap analysis and migration roadmap.
patternsDiagnose problems and analyze architectures against agentic design patterns.
project-auditFull audit orchestrator that runs applicable audits and produces docs/audits/ output.
security-auditSecurity audit across auth, input validation, LLM-specific attacks, data access, APIs, secrets, and infrastructure.
specAdversarial spec interview for long autonomous runs, including takt/factory handoff material.

Productivity

SkillDescription
orchestratorSenior-architect delegation mode for decomposing non-trivial work, routing worker agents, and reviewing output.
uiGenerate complete HTML+CSS pages in International Typographic Style.
componentGenerate self-contained UI components such as buttons, cards, modals, and tables.
pageGenerate full page layouts such as landing, pricing, dashboard, and blog pages.
refineIterate on generated POCs until the design is approved.
prdConvert approved UI POCs into implementation-ready Product Requirements Documents.
product-contextGuided interview that captures product identity, value proposition, aha moment, and user journey.
brand-voiceGuided interview that produces voice traits, tone map, customer language inventory, and competitive positioning.
ux-auditSaaS UX audit that scans user-facing touchpoints and applies established UX frameworks.
ux-copyWrites production-ready UX copy for user-facing surfaces using microcopy and CTA frameworks.

Misc

SkillDescription
client-proposalGenerate client-facing proposals with a built-in review for internal leaks.
debugTrack a debugging session with a breadcrumb file so work can be resumed without losing context.
skill-reviewReview Claude Code skill files for format, interaction design, chain integration, and effectiveness.

Reference Docs

Reference docs live in docs/ and are installed by ./install.sh.

DocUsed By
docs/agentic-patterns/patterns-master-summary.mdpatterns, agent-review, agent-audit, harness-architect
docs/agentic-patterns/patterns-selection-guide.mdpatterns, agent-review, harness-architect
docs/agentic-patterns/patterns-anti-patterns.mdpatterns, agent-review, agent-audit, harness-architect
docs/harness-patterns/harness-types.mdharness-architect, harness-review
docs/harness-patterns/harness-selection-guide.mdharness-architect, harness-review
docs/skill-authoring-guide.mdharness-build, skill-review, skill authors
docs/deep-review/framework.mddeep-review
docs/deep-review/resources.mddeep-review
docs/deep-review/security.mddeep-review
docs/deep-review/absence.mddeep-review
docs/deep-review/research-dimensions.mddeep-review
docs/skill-stewardship-module.mdSkill stewardship module design and non-goals

Platforms

Claude Code is the supported platform today. The repo includes a first platform manifest and adapter docs as a starting point for targeting Claude Code, Codex, OpenCode, and Pi-mono.

FilePurpose
manifest/platform-matrix.yamlDeclares supported/planned platforms, install targets, artifact types, and retired repo-managed artifacts.
platforms/claude-code/README.mdDocuments the current Claude Code install paths and compatibility rules.

Modules

Modules are optional infrastructure around the agent/skill system. They are not installed by default. They must stay advisory unless a user explicitly opts into stronger behavior.

ModulePurpose
modules/skill-stewardship/Advisory skill lifecycle module for suggesting new skills, improvements, splits, redirects, retirements, and later verification.
platforms/claude-code/skill-stewardship/Claude Code hook wiring notes for the Skill Stewardship module.

Installation

git clone git@github.com:Sebstrdigital/agentic-tools.git
cd agentic-tools
./install.sh

The installer copies:

  • agents/*.md -> ~/.claude/agents/
  • docs/*/*.md -> matching ~/.claude/docs/<subdir>/
  • docs/*.md -> ~/.claude/docs/
  • skills/<category>/<skill-name>/*.md -> ~/.claude/skills/<skill-name>/

It also removes repo-managed legacy command files from ~/.claude/commands/ after their source files migrate to skills. User-owned commands are left untouched.

Safe install behavior:

  • If a file with the same name already exists and came from this repo (source_id: seb-claude-tools), it gets updated with version tracking.
  • If a file with the same name exists but belongs to something else, it installs with a sebstrdigital- prefix to avoid conflicts.
  • If no conflict exists, it installs directly.

After installing, restart Claude Code to pick up changes.

Updating

Pull the latest and re-run the installer:

cd agentic-tools
git pull
./install.sh

The installer shows version changes, for example v1.0.0 -> v1.1.0, and skips files that are already up to date.

File Structure

agentic-tools/
├── agents/
│   ├── builder.md
│   ├── grunt.md
│   ├── heavy.md
│   ├── scout.md
│   └── skeptic.md
├── docs/
├── manifest/
│   └── platform-matrix.yaml
├── modules/
│   └── skill-stewardship/
├── platforms/
│   └── claude-code/
│       ├── README.md
│       └── skill-stewardship/
├── skills/
│   ├── engineering/
│   ├── misc/
│   └── productivity/
├── CLAUDE.md
├── install.sh
└── README.md

Metadata

Repo-managed agent, doc, and skill files include YAML frontmatter with:

  • source_id: seb-claude-tools to identify files from this repo.
  • version: 1.0.0 to track updates.

The installer uses this metadata to update repo-managed files in place while avoiding unmanaged local files.

License

MIT

Ver o original no GitHub

Bom para

Relacionados