Back to directory

Microsoft Security Skills Plugin

A curated set of 88 Microsoft security skills for AI coding assistants, covering Defender, Sentinel, Entra, and Purview, so an agent gives grounded, product-specific guidance instead of generic advice.

Microsoft Security Skills Plugin

The Microsoft Security Skills Plugin packages Microsoft security expertise as installable skills for existing AI coding assistants, including GitHub Copilot, Claude Code, Cursor, Codex CLI, and Gemini CLI, rather than shipping as its own standalone tool. It covers 88 skills across threat protection and security operations, such as Defender for Endpoint, Defender for Identity, and Sentinel; identity, access, and governance, such as Entra ID, conditional access, and Azure PIM; and compliance and data protection, such as Purview data-loss-prevention policy.

Each skill provides workflows, decision trees, and guardrails grounded only in public Microsoft Learn documentation, aiming to help an agent decide which control applies in a given situation, what needs validating before a policy change goes live, and which investigation path or guardrail matters, rather than giving generic security advice. The project states each skill's measured benefit over an unaided model has been validated with a reproducible harness across two independent frontier models.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.