Kavach Skill
SarthiAI/kavach-skill
Segurança aplicada: gating, drift detection e tokens pós-quânticos para agents
Ver no GitHubSkill oficial para integrar Kavach (gating de segurança, detecção de drift, tokens assinados e políticas deny-by-default) em serviços Python e Node/TypeScript. Fornece guia prático para instalação do SDK, autoria de políticas, drift detection, audit chains e secure channels.
Instalação
npx convez-ai SarthiAI/kavach-skill1 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
- Adicionar policies de controle de acesso a serviços
- Implementar detecção de anomalias (device, geo, session age, action count)
- Emitir e validar tokens assinados pós-quânticos
- Manter audit chains tamper-evident
- Hot-reload de políticas e rollout gradual em modo observação
Veredito

Skill técnico e maduro para quem precisa de segurança declarativa em produção. Cobre tanto Python quanto Node com a mesma semântica, suporta múltiplos formatos de policy e inclui padrões avançados (tokens pós-quânticos, secure channels). Requer familiaridade com políticas de controle de acesso.
README
A tradução para o português deste README ainda está na fila.
Kavach Skill
The official skills library for Kavach execution gates, covering both the Python and Node / TypeScript SDKs.
What it does
When you ask your coding agent to add Kavach gating, drift detection, signed permit tokens, or default-deny policy enforcement to a service, this skill activates and walks the agent through the correct setup. The skill ships parallel guidance for Python (kavach-sdk on PyPI) and Node / TypeScript (kavach-sdk on npm), backed by the same Rust core.
Install
npx skills add SarthiAI/kavach-skill
That places kavach/ into your project's skills directory. Compatible clients pick it up automatically.
What's covered
- Installing
kavach-sdk(PyPI or npm) and constructing aGate. - Authoring policies as a Python dict, TypeScript object, JSON string, JSON file, TOML string, or TOML file (one schema, five loaders, both languages).
- Wiring the four built-in drift detectors (device, geo, session age, action count) in either binding.
- Issuing signed post-quantum permit tokens with
PqTokenSignerand verifying them downstream. - Maintaining a tamper-evident
SignedAuditChainand exporting it as JSONL. - Hot-reloading policies and rolling out gradually with observe-only mode.
- Building encrypted + signed + replay-protected byte channels with
SecureChannel.
Layout
kavach/
├── SKILL.md concepts + capabilities + language pointers
├── assets/
│ └── policies.example.toml shared kitchen-sink TOML policy
├── references/
│ └── policy-language.md shared condition grammar
├── python/ Python integration: README, sdk, drift, audit, secure-channel, scaffold
└── node/ Node integration: README, sdk, drift, audit, secure-channel, scaffold
The agent reads SKILL.md first to understand what Kavach does, then loads python/README.md or node/README.md depending on the user's stack.
Scope
This skill covers the published kavach-sdk package on PyPI and npm: the Gate class (with all five loaders), action context / evaluate options, Verdict, PermitToken, the four built-in drift detectors, signed permit tokens, the audit chain, the public-key directory, SecureChannel, the Python @guarded decorator, and the Node MCP middleware guardTool. Anything outside that surface is not in scope and the skill should not suggest it. For HTTP / MCP middleware deep dives or multi-replica Redis deployments, see the Kavach roadmap.
License
Apache-2.0. The skill content (Markdown, example policies, scaffold scripts) is freely embeddable in any agent runtime. The Kavach library itself is released under Elastic License 2.0; see that repo for terms governing the library.