convez.ai
AgenteSegurançaAvançado
tayontech

SCOPE - Purple Team de Segurança Cloud

tayontech/SCOPE

Audita AWS, valida ataques, gera playbooks de exploit e guia investigações

Ver no GitHub
BomPluginScore · de 0 a 100

Agent IA especializado em auditoria de segurança AWS com workflows de purple teaming determinísticos. Executa enumeração de recursos, valida caminhos de ataque, gera playbooks de exploit (apenas revisão) e guia investigações em Splunk. Roda nativamente em Claude Code, Gemini CLI e outros ambientes de código assistido.

Instalação

$npx convez-ai tayontech/SCOPE

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
54
Forks
12
Acompanhando
3
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
1
Contribuidores
1

Casos de uso

  • Auditoria de segurança em ambientes AWS multi-conta
  • Validação de caminhos de ataque e exploração controlada
  • Geração de controles e regras de detecção Splunk
  • Investigação guiada de alertas de segurança

Veredito

Ferramenta sólida para times de segurança ofensiva em AWS, com controles bem pensados (read-only por padrão, hooks de segurança, artifacts apenas para revisão). Excelente para purple teams que precisam automatizar auditorias e validar exposições reais sem risco de execução acidental.

README

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

SCOPE: Security Cloud Ops Purple Engagement

GitHub stars GitHub forks License: MIT GitHub last commit

SCOPE runs AWS purple-team workflows with deterministic Python inventory and bounded AI agents. It audits AWS resources, validates attack paths, generates review-only exploit playbooks, produces controls, and guides Splunk investigations.

How It Works

Run SCOPE from Claude Code, Antigravity CLI, Gemini CLI, or Codex CLI:

/scope:audit --all
PhaseOutput
AuditPython enumerators inventory IAM, STS, S3, KMS, Secrets Manager, Lambda, EC2, ECS, RDS, API Gateway, SNS, SQS, CodeBuild, Bedrock, CloudFront, Cognito, DynamoDB, Route 53, and SSM
Attack pathsCandidate generation, validation, grouped reporting, and review-only AWS CLI replay artifacts
ControlsOrg-wide issues, SPL detections, monitoring dashboard ideas, policy replacements, remediation, and validation
ExploitPrincipal-scoped playbooks with approved replay command artifacts
InvestigateAlert, run-guided, and intel-driven Splunk investigation workflows

Quick Start

git clone https://github.com/tayontech/SCOPE.git
cd SCOPE
uv run python -m scope.install

export AWS_PROFILE=your-profile

/scope:audit --all
/scope:audit iam s3 lambda
/scope:exploit arn:aws:iam::123456789012:role/target-role
/scope:exploit
/scope:investigate

Requirements: Python 3.11+, uv, AWS CLI with read-only credentials, and one supported runtime. Node.js supports dashboard generation and Splunk MCP mcp-remote transport.

Scripted installs:

uv run python -m scope.install --claude --local --no-splunk-mcp
uv run python -m scope.install --antigravity --local --no-splunk-mcp
uv run python -m scope.install --gemini --local --no-splunk-mcp
uv run python -m scope.install --codex --local --no-splunk-mcp
uv run python -m scope.install --all --local --with-splunk-mcp

Interactive install asks whether to configure bundled Splunk MCP defaults. Scripted installs skip MCP server settings unless you pass --with-splunk-mcp. Use --no-splunk-mcp when automation should show the choice in logs. SCOPE remains SPL-first until the project adds query-language profiles. See config/mcp-setup.md.

Safety

SCOPE agents run read-only AWS activity by default. config/hooks/scope-safety-guard.sh blocks destructive AWS shell operations before execution. Audit and exploit may write AWS CLI replay command artifacts for human review; agents do not execute those write commands.

HookPurpose
Safety GuardBlocks destructive AWS shell operations
SPL LintRejects SPL anti-patterns and side-effect commands
Schema ValidateValidates results and module envelopes
Artifact CheckChecks required run outputs before completion

Dashboard

Generate one self-contained report for a run:

cd dashboard && npm run dashboard
open dashboard/reports/<run-id>-dashboard.html

An explicit audit run directory basename becomes the dashboard run ID. dashboard/public/index.json stores reports[]; each report points to one audit JSON export and optional controls JSON export. Controls attach to the audit report, so one audit workflow creates one selectable dashboard report.

The dashboard shows attack graphs, path details, public exposure findings, controls, SPL detections, dashboard ideas, policy replacements, and remediation.

Platforms

PlatformStatusHook ConfigInstall Surface
Claude CodeSupported.claude/settings.json.claude/skills/, .claude/agents/
Antigravity CLIPreferred Google target.agents/hooks.json.agents/skills/, .agents/mcp_config.json, .agents/plugins/scope/agents/
Gemini CLILegacy Google target.gemini/settings.json.agents/skills/, .gemini/agents/
Codex CLISupported.codex/hooks.json.agents/skills/, .codex/agents/

Google announced on May 19, 2026 that Gemini CLI and Gemini Code Assist IDE extensions stop serving requests for Google AI Pro, Ultra, and free individual users on June 18, 2026. Use Antigravity CLI for new Google installs. SCOPE keeps --gemini for enterprise/API-key users and migration.

Agents And Models

Top-level agents inherit the runtime session model:

  • scope-audit orchestrates audit gates, Python runtime execution, attack analysis, replay artifacts, and controls chaining.
  • scope-controls orchestrates org-wide issues, detections, dashboard ideas, policy replacements, remediation, and validation.
  • scope-exploit generates principal-scoped red-team playbooks.
  • scope-investigate runs alert, run-guided, and intel investigation modes.

Subagents run bounded reasoning tasks:

  • scope-attack-analyze, scope-attack-validate, scope-public-exposure-analysis, and scope-awscli-replay support attack-path analysis and replay artifacts.
  • scope-controls-org-wide, scope-controls-detections, scope-controls-dashboards, scope-controls-policy, scope-controls-remediation, and scope-controls-validate support controls.
  • scope-investigate-alert, scope-investigate-intel, and scope-investigate-run prepare investigation context.
  • scope-research - shared external technique research for attack analysis and exploit playbooks.

When you run /scope:audit --all, the orchestrator runs on your session model, calls scope audit for deterministic Python enumeration and post-processing, seeds IAM and public/service-connected candidates with scope.attack.candidates, dispatches scope-attack-analyze, optionally enriches candidates through scope-research, then can generate review-only AWS CLI replay artifacts for validated paths before chaining controls on a reasoning model.

RuntimeReasoning Subagent Tier
Claude Codeopus[1m] alias
Antigravity CLIModel selected in Antigravity
Gemini CLIpro alias
Codex CLIgpt-5.5 with high reasoning effort

Claude subagents use the opus[1m] alias for larger context during artifact-heavy analysis and validation. Enumeration uses deterministic Python, not an AI model.

Documentation

DocumentPurpose
ARCHITECTURE.mdComponent ownership, pipeline flow, runtime contracts, graphs, and hooks
RELEASE-NOTES.mdCurrent release changes
config/mcp-setup.mdSplunk MCP and manual SPL mode
config/README.mdConfig ownership
knowledge/README.mdDurable knowledge rules and redaction
config/project-docs/PROJECT.mdSource for generated runtime instruction files
docs/LLM-CONTEXT.mdReviewer and implementation-agent context
dashboardReact and D3 report generator
config/hooksSafety and validation hooks
config/schemasJSON Schema contracts

Community

Created by Tayvion Payton

Ver o original no GitHub

Bom para

Relacionados