Ricar66/claude-agents

English · Português
A collection of subagents for Claude Code, each with a ready-to-install agent, documentation, evaluations and a way to receive your organization's context without that context entering the repository.
## Agents
| Agent | What it does | Status |
| security-sentinel | Read-only, evidence-based security audit: code, APIs, network (via files), infrastructure, supply chain, CI and AI systems (LLM, agents, MCP, RAG). Eleven modes, from triage to release-gate | Initial (v0.1) |
| infra-guardian | Read-only infrastructure and reliability review (SRE, DevOps, platform, databases, FinOps): VPS, Docker, Kubernetes, IaC, network, CI/CD, observability, backup and DR, capacity, incidents and cost. Seventeen modes, from discover to verify-change; it never applies changes, it delivers plans with rollback. Comes with a static variant without Bash | Initial (v0.1) |
| backend-architect | Backend architect and engineer: APIs, domain rules, databases, integrations, migrations and tests. Reads the whole repository before proposing, preserves what works and implements when asked. Fifteen modes, from discover to verify | Initial (v0.1) |
| frontend-architect | Frontend architect and engineer: UX, design system, accessibility (WCAG 2.2 AA), responsiveness, rendering, performance and tests. Follows the existing visual identity and implements when asked. Seventeen modes | Initial (v0.1) |
The agents complement each other: the two auditors (Sentinel and Guardian) only read; the two architects design and, if you ask, implement. How to use and call each one, and how they work together: GUIDE.md.
## How each agent is organized
text
├── agent/
│ ├── pt-BR/ .md ← copy ONE language to ~/.claude/agents/
│ └── en/ .md (some agents also ship variants, e.g. -static.md)
├── org-context-example/ ← fictitious template of the organizational context
│ ├── pt-BR/SKILL.md
│ └── en/SKILL.md
└── docs/
├── pt-BR/
└── en/
## Organizational context stays on your machine
A generic agent makes more mistakes. That is why each agent can preload a local skill, -org-context, with what only your organization knows: systems, where they run, network, accounts, known risks and internal rules.
The skill lives at ~/.claude/skills/ -org-context/SKILL.md, outside this repository.
If it does not exist, Claude Code skips the preloading and the agent runs in generic mode.
This repository's .gitignore ignores -org-context/ folders as a safeguard, but the main rule is: real context never goes in here.
Never put a password, token or key in the context. Write where the credential is and what the risk is.
## Quick install
powershell
# Windows (PowerShell)
Copy-Item .\security-sentinel\agent\en\security-sentinel.md "$env:USERPROFILE\.claude\agents\"