cloud-security-agent-skills
vaquarkhan/cloud-security-agent-skills
it's a real agent harness with cloud-provider modules (AWS, Azure, GCP, OCI, IBM, Alibaba, PCF), a bastion security gateway, and an MCP server, plus 14 skills.
Instalação
npx convez-ai vaquarkhan/cloud-security-agent-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
README
A tradução para o português deste README ainda está na fila.
cloud-security-agent-skills
Zero-trust multi-cloud security for AI agents — 14 end-to-end Agent Skills, a real MCP-Bastion stdio proxy, IdentityManager with ephemeral credentials for 7 clouds, Well-Architected framework mappings (AWS · Azure · GCP · OCI · IBM Bluemix · Alibaba · PCF/Tanzu), plus VS Code and JetBrains install plugins.
Disclaimer: Operational security patterns and automation templates — not a substitute for your cloud QSA, CISO sign-off, or penetration test report.
Coverage: IAM, KMS/vault, network, compute/K8s, logging, detection, compliance, and agent/MCP deployment for every major cloud. Provenance trail in registry/provenance.yaml. Full status: docs/project-status.md.
Why this exists
Security agents must not:
- Embed static API keys or long-lived tokens in code or MCP configs
- Connect to MCP tools without a security gateway (PII redaction, injection defense)
- Review AWS IAM while ignoring Azure Entra ID or GCP Workload Identity in hybrid estates
- Ship skills without validation, provenance, or IDE install surfaces
cloud-security-agent-skills delivers:
- 14 specialized skills (~215 lines avg) — deepest end-to-end cloud security content in the agent-skills family
- IdentityManager — thread-safe ephemeral credentials (AWS IRSA, Azure MI, GCP WIF, OCI resource principal, IBM Trusted Profile, Alibaba RAM, PCF CredHub)
- MCP-Bastion gateway — real
bastion_proxy.pystdio proxy;bastion.yamlpolicy-as-code (Presidio, PromptGuard, RBAC, rate limits) - Well-Architected one-stop shop — six pillars mapped across all clouds including Bluemix and Pivotal PCF legacy names
- IDE plugins — VS Code extension + JetBrains plugin (install skills, presets, MCP templates, adapters)
- Governance — validators, benchmark, AIV PR gate, CodeQL, Dependabot, pre-commit
Quick start
Prerequisites
- Python 3.10+
- Node.js 18+ (VS Code extension build; optional)
- JDK 17+ (JetBrains plugin build; optional)
Bootstrap (recommended)
macOS / Linux:
git clone git@github.com:vaquarkhan/cloud-security-agent-skills.git
cd cloud-security-agent-skills
chmod +x bootstrap.sh scripts/install.sh
./bootstrap.sh
Windows (PowerShell):
git clone git@github.com:vaquarkhan/cloud-security-agent-skills.git
cd cloud-security-agent-skills
.\bootstrap.ps1
Detects your workflow and installs skills, Cursor rules, Claude commands, Copilot instructions, MCP templates, and hooks.
Run the agent
pip install -e ".[dev]"
cloud-security-agent
python examples/mock-posture-check/run_posture_check.py
Validate the repository
make validate
make test
make demo
Plugin installation
Cursor
./bootstrap.sh --target cursor
See docs/cursor-setup.md — installs .cursor/rules/ + skills.
VS Code / GitHub Copilot
cd vscode-extension && npm install && npx @vscode/vsce package
code --install-extension cloud-security-agent-skills-*.vsix
Or: ./bootstrap.sh --target vscode
See docs/vscode-setup.md · vscode-extension/README.md
Claude Code
./bootstrap.sh --target claude
Copies .claude/commands/ and CLAUDE.md.
JetBrains (IntelliJ, PyCharm, …)
cd jetbrains-plugin && ./gradlew buildPlugin
Tools → Cloud Security Agent Skills → Install Full Toolkit
See docs/jetbrains-setup.md · jetbrains-plugin/README.md
Security lifecycle commands
| Command | Purpose | Skill / doc |
|---|---|---|
/posture-check | Skills complete, Bastion valid, routing | .claude/commands/posture-check.md |
/aws-security-review | Full AWS IAM/KMS/network review | .claude/commands/aws-security-review.md |
| Assess | Posture + WAF pillar scoring | cloud-well-architected-frameworks |
| Harden | Remediate IAM/KMS/network gaps | Per-cloud skill |
| Validate | mcp-test security suite | mcp-security-testing-harness |
| Gate | AIV PR integrity | pr-integrity-aiv-gate |
Example flow:
/posture-check → load cloud skill → WAF review → mcp-test → AIV gate
Skills catalog
14 skills — full index: skills-index.md
| Group | Skills |
|---|---|
| Meta | using-cloud-security-agent-skills, cloud-well-architected-frameworks, zero-trust-identity-and-secrets, multi-cloud-security-posture |
| Cloud | aws, azure, gcp, oci, ibm (Bluemix), alibaba, vmware-tanzu-pcf (Pivotal CF) |
| Tooling | mcp-bastion-security-gateway, mcp-security-testing-harness, pr-integrity-aiv-gate |
Each cloud skill: governance → IAM → KMS → network → compute → data → logging → compliance → checklist → red flags.
Cloud presets & starter packs
| Cloud | Preset | Starter pack |
|---|---|---|
| AWS | presets/aws-security/ | aws-agent-starter.yaml |
| Azure | presets/azure-security/ | azure-agent-starter.yaml |
| GCP | presets/gcp-security/ | gcp-agent-starter.yaml |
| OCI | presets/oci-security/ | — |
| IBM / Bluemix | presets/ibm-bluemix-security/ | — |
| Alibaba | presets/alibaba-security/ | — |
| PCF / Tanzu | presets/tanzu-pcf-security/ | — |
| Multi-cloud | presets/multi-cloud-hybrid/ | multi-cloud-agent-starter.yaml |
Well-Architected frameworks
One-stop cross-cloud WAF: skills/cloud-well-architected-frameworks/SKILL.md
Full pillar matrix: well-architected-pillar-matrix.md
| Provider | Framework |
|---|---|
| AWS | AWS Well-Architected Framework |
| Azure | Microsoft Azure Well-Architected Framework |
| GCP | Google Cloud Architecture Framework |
| OCI | OCI Architecture Center |
| IBM | IBM Cloud / Bluemix + SCC |
| Alibaba | Alibaba Cloud Well-Architected |
| PCF | Tanzu Reference Architecture + NIST CSF |
Project structure
cloud-security-agent-skills/
├── skills/ # 14 Agent Skills (SKILL.md each)
├── security/ # IdentityManager — 7 clouds
├── agent/ # CloudSecurityAgent, Bastion proxy, MCP server
├── presets/ # 8 cloud WAF/security presets
├── starter-packs/ # YAML starter bundles
├── examples/mock-posture-check/
├── vscode-extension/ # VS Code / Cursor / Copilot plugin
├── jetbrains-plugin/ # IntelliJ-family plugin
├── registry/
│ ├── assets.json # Machine-readable index
│ ├── install-manifest.json # Plugin install manifest
│ └── provenance.yaml # SME source trail
├── scripts/ # validate-skills, validate-assets, install
├── evals/benchmark/ # Skill routing benchmark (100% target)
├── hooks/ # Session + static-secret guards
├── templates/ # Bastion MCP client, posture checklist
├── references/ # Multi-cloud control matrix
├── mcp/ # Bastion MCP client template
├── docs/ # Setup guides per IDE
├── .cursor/rules/ # Cursor agent rules
├── .claude/commands/ # Claude slash commands
├── bastion.yaml # MCP-Bastion policy
└── AGENTS.md # Agent routing (read first)
Full map: docs/folder-structure.md
MCP integrations
Bastion-wrapped MCP — never connect directly to agent.mcp_server:
| Template | Use case |
|---|---|
| mcp/cloud-security-agent.mcp.json | Cursor / Claude Desktop stdio |
| templates/bastion-mcp-client.json | Copy-paste client config |
Tools: assess_cloud_posture, validate_security_controls, echo_security_context
See mcp/README.md and skill mcp-bastion-security-gateway.
Examples
| Example | Description |
|---|---|
| examples/mock-posture-check/ | Multi-cloud posture check — no cloud credentials required |
python examples/mock-posture-check/run_posture_check.py --cloud aws --cloud azure --json
Documentation
| Doc | Topic |
|---|---|
| docs/project-status.md | Feature list and version summary |
| docs/architecture.md | Threat model + data flow |
| docs/getting-started.md | First-run walkthrough |
| docs/testing.md | Unit + MCP harness + local dev |
| docs/cursor-setup.md | Cursor install |
| docs/vscode-setup.md | VS Code / Copilot install |
| docs/jetbrains-setup.md | JetBrains plugin |
| docs/plugin-publishing.md | Release VSIX / plugin ZIP |
| docs/sme-review.md | Provenance cadence |
| AGENTS.md | Agent entry + routing |
| CONTRIBUTING.md | Development workflow |
| SECURITY.md | Vulnerability disclosure |
Identity (zero-trust)
| Cloud | Mechanism | Never use |
|---|---|---|
| AWS | IRSA / instance profile | Access keys |
| Azure | Managed identity | Client secret on compute |
| GCP | Workload Identity | Downloaded SA JSON |
| OCI | Resource principal | API signing key in code |
| IBM / Bluemix | Trusted Profile | Service ID API key |
| Alibaba | ECS RAM role / RRSA | AccessKey in env |
| PCF / Tanzu | UAA + CredHub | Secrets in manifest.yml |
License
MIT — see LICENSE.
Acknowledgments
Built on MCP-Bastion, mcp-test-harness, aiv-integrity-gate, and the Agent Skills progressive-disclosure pattern. Packaged like compliance-agent-skills for family consistency.