Installs into .claude/skills of the current project.
Sponsors
SSkills DirectorySkills Directory
Know which skills are safe — weekly.
SSkills DirectorySkills Directory
Is that skill safe to install?
Are you the author of Secure Architecture?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/andreibesleaga-secure-architecture)
---
name: secure-architecture
description: "Architect systems with defense-in-depth and Zero Trust principles. Use when: design secure system, enforce zero trust, audit architecture security."
license: CC-BY-SA-4.0
metadata:
source: "agents/skills/security/secure-architecture.skill.md"
tags: "security"
core: "false"
context_cost: "high"
---
# secure-architecture
## Goal
Enforce robust, defense-in-depth architectural patterns including Zero Trust boundaries, secure access service edge (SASE), network segmentation, and strict IAM.
## Steps
1. **Perimeter Audit**: Assume breach. Establish micro-perimeters around critical logical domains.
2. **Access Policy**: Design the Authorization logic (e.g., using OPA or AWS IAM) ensuring least privilege access across all services.
3. **Data Protection**: Mandate explicit data transit TLS boundaries and rest encryption.
4. **Verification**: Verify that the C4 diagrams explicitly mark trusted vs untrusted zones and the security boundary implementations.
## Security & Guardrails
### 1. Skill Security
- **Config Drift Ban**: The agent must not manually execute cloud IAM changes without outputting them as declarative Infrastructure as Code (e.g., Terraform/CloudFormation) for review.
- **Root Veto**: Explicitly forbid creating or assigning "Admin", "*", or equivalent root-level global actions to any service role inside the generated documentation.
### 2. System Integration Security
- **MFA Enforcement**: Any external/remote access vectors defined in the architecture must explicitly mandate Multi-Factor Authentication.
- **Fail Closed Mechanism**: All network authorization services, if unavailable, must fail-closed (denying access), never fail-open.
### 3. LLM & Agent Guardrails
- **Trust Assumption Bias**: The LLM must not implicitly trust an internal network. Every node must re-validate the identity and authorization of the caller (Zero Trust).
- **Compliance Illusion**: The LLM cannot authorize itself or the architecture as "HIPAA Compliant" or "SOC2 Compliant" because it enforces Zero Trust; formal auditor attestation remains legally required.