Agent Red Team

Run an adversarial assurance review of a Microsoft Copilot Studio agent you own — generate and execute test cases for indirect prompt injection, oversharing and authorization bypass, data leakage, scope escape and tool misuse — then score the findings and map each one to a concrete Copilot Studio control. Use when someone asks to red team, pen test, adversarially test, jailbreak-test or security-review an agent, wants a prompt injection or data leakage assessment before go-live, or needs security sign-off and a risk report for an agent handling sensitive data.

110 stars 122 forksSecurityOfficial

Install

skills CLI (Claude Code, Cursor, Codex and more)
npx skills add microsoft/cat-agent-skills --skill agent-red-team
Manual install
git clone --depth 1 https://github.com/microsoft/cat-agent-skills.git /tmp/cat-agent-skills
mkdir -p ~/.claude/skills
cp -r /tmp/cat-agent-skills/submissions/agent-red-team ~/.claude/skills/agent-red-team

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# Agent Red Team

Functional test plans ask whether the agent does the right thing for a cooperative user.
This asks what it does for an uncooperative one — and, more importantly in practice, for
a cooperative user reading a document somebody else poisoned.

The dominant real-world risk for a grounded enterprise agent is not a clever user typing
a jailbreak. It is **indirect prompt injection**: instructions hidden inside content the
agent retrieves, and oversharing, where the agent faithfully surfaces documents the asker
was never meant to see. Weight the review accordingly.

## Authorization gate — do this first, every time

Before generating a single test case, establish and record in the report:

1. **Ownership.** The user owns the agent, or has written authorization from the owner to
   test it. If neither, stop. Say plainly that adversarial testing of someone else's agent
   needs their authorization, and offer to help them request it.
2. **Environment.** Test in dev or a dedicated test environment wherever possible. If
   production is the only option, get that in writing and note it as a finding in itself.
3. **Data handling.** Use **synthetic canary data** — never real personal data, real
   credentials or real customer records — to test leakage. A canary is a distinctive
   fabricated string planted in a document the agent can reach; if it appears in output
   where it should not, you have proof without exposing anything real.
Read the full skill on GitHub

Details

Skill name
agent-red-team
Path
submissions/agent-red-team/SKILL.md
License
MIT
Repo updated
15 November 2026

FAQ

What does the Agent Red Team skill do?

Run an adversarial assurance review of a Microsoft Copilot Studio agent you own — generate and execute test cases for indirect prompt injection, oversharing and authorization bypass, data leakage, scope escape and tool misuse — then score the findings and map each one to a concrete Copilot Studio control. Use when someone asks to red team, pen test, adversarially test, jailbreak-test or security-review an agent, wants a prompt injection or data leakage assessment before go-live, or needs security sign-off and a risk report for an agent handling sensitive data.

How do I install the Agent Red Team skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add microsoft/cat-agent-skills --skill agent-red-team. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Agent Red Team popular?

It lives in microsoft/cat-agent-skills, which has 110 GitHub stars and 122 forks.

Related skills

3 Statements

110

microsoft/cat-agent-skills

Completes existing income statement, balance sheet and cash flow templates while preserving layout, grounding historical inputs and linking forecast formulas with reconciliation checks. Use for "fill in a 3-statement template", "complete an IS/BS/CF framework", "populate my integrated financial model", "link statements in this template", or "finish the forecasts in my three-statement workbook". Do NOT use for blank-slate models or general workbook edits (use xlsx), DCF valuation (use dcf-valuation), or buyout modeling (use lbo-model).

AI, LLMs & Agents122

Accessibility Pass

110

microsoft/cat-agent-skills

Use this skill whenever the user asks to check, review, or fix the accessibility of a PowerPoint deck, Word document, HTML page, or Markdown file, and before handing over any deck or document this agent just generated, so it does not ship with missing alt text, untitled slides, or unreadable colour contrast.

Frontend & Design122

Accessible Learning Materials

110

microsoft/cat-agent-skills

Maintains instructionally equivalent, accessible variants of a set of learning materials — lessons, readings, slides, and handouts — with parity checks against the original learning goals and recorded approvals. Use when the user asks to "make these materials accessible", "audit this lesson set for accessibility", "create an accessible version of this reading", "check reading level and alt text", or "keep my accessible variants in sync". Do NOT use for individual student accommodation or support-plan decisions (IEP, EHCP, or local equivalent), and do NOT use for generic document formatting (us

Learning & Education122

Exam Request Response

110

microsoft/cat-agent-skills

Turns a regulator or auditor request list, PDL, IRL, or first-day letter into a tracked response register with original numbering, owners, candidate evidence, status, gaps, and a draft memo. Never asserts that evidence satisfies a request and never sends anything. Use when the user says "build a tracker for this PDL", "who owns each exam item", "find evidence for this request list", "draft responses to the first-day letter", or "what are we missing for the exam". Do NOT use for vendor SOC 2 or DDQ reviews (use vendor-risk-review), enforcement read-across (use enforcement-action-review), one po

Security122

Soc Stakeholder

248

asgard-ai-platform/skills

Conduct stakeholder analysis using identification, Power-Interest matrix classification, and influence strategy development. Use this skill when the user needs to map stakeholders for a project, manage conflicting interests, prioritize communication, or build a stakeholder engagement plan — even if they say 'who needs to approve this', 'how do I get buy-in', or 'who might block this project'.

Security30

Power Automate Desktop Assessment

110

microsoft/cat-agent-skills

Assess Power Automate Desktop and hybrid cloud/desktop automation projects for maintainability, reliability, security, performance, troubleshooting readiness, scaling, and governance using current Microsoft guidance and evidence-based review practices.

Security122

Browse the most popular Claude skills, all security skills, or write your own with the skills generator.