| name | securing-mas |
| description | Apply OWASP MAESTRO, MITRE ATLAS, NIST AI RMF, and ISO 42001/23894 security frameworks to MAS designs |
| compatibility | Designed for Claude Code |
| metadata | {"allowed-tools":"Read, Grep, Glob, WebSearch, WebFetch","argument-hint":["component-or-feature"],"stability":"stable","content-hash":"sha256:94039e17e94bb06a07cb298b68d4b5a02f3ef882ad08837656d1a82d3e0ad8b2","last-verified-cc-version":"1.0.34"} |
Securing Multi-Agent Systems
Target: $ARGUMENTS
When to Use
Trigger this skill when:
- Conducting security reviews of agent systems
- Threat modeling for multi-agent architectures
- Reviewing plugin implementations for security
- Designing security controls for pipelines
Framework Stack
MITRE ATLAS (attack taxonomy — what adversaries do)
| informs threat identification
v
OWASP MAESTRO (threat model — what to defend against in MAS)
| maps threats to controls
v
NIST AI RMF (risk framework — how to govern/map/measure/manage)
| operationalized by
v
ISO 42001 + 23894 (certifiable management system + risk methodology)
Use all four layers together: ATLAS enumerates attack vectors, MAESTRO maps
them to MAS-specific controls, NIST AI RMF structures governance, and ISO
provides the certifiable management system.
Workflow
-
Review the framework stack — references/mas-security.md for the conceptual overview of MAESTRO, ATLAS, NIST AI RMF, and ISO 42001/23894 layers working together.
-
Apply the 7-layer security check — for each new component, walk through every MAESTRO layer. See references/maestro-7-layer-checklist.md for the actionable per-layer checklist (Model → Orchestration).
-
Run the plugin security checklist — before marking an implementation complete, verify input validation, output safety, resource management, observability, and external dependencies. See references/plugin-security-checklist.md.
-
Document threats in the cross-framework matrix — for each feature, map concerns to ATLAS techniques, MAESTRO layers, NIST functions, and ISO controls. Start from references/threat-matrix-template.md and add feature-specific rows.
-
Avoid common vulnerability patterns — consult references/common-vulnerabilities.md for vulnerable/secure code examples: prompt injection (L1), type confusion (L2), resource exhaustion (L5), secret leakage (L6).
-
Test security controls explicitly — write tests that exercise each MAESTRO layer's controls. See references/security-testing-patterns.md for pytest examples (input validation, timeout enforcement, error message safety).
Further Reading
