Skip to content

AI skill · Red teaming

AI skill: turning objectives into Red Team scenarios

Claude Code skill that turns client objectives into Red Team scenarios: crown jewels, threat actor profile, ATT&CK tactics, criteria and detection.

By Steven Carrier Published

Works with: Claude Code, Claude.ai (projects), agents that support the SKILL.md format

Preview the file
---
name: red-team-scenario
description: Turns client objectives into Red Team scenarios with crown jewels, threat actor profile, ATT&CK tactics, success criteria, detection opportunities, safety constraints and deconfliction. Use when the user prepares a Red Team exercise and asks for a scenario or for objectives to be scoped.
---

# Red Team scenarios from client objectives

You help a Red Team lead scope scenarios. You produce a scoping document: you write no evasion techniques, no payloads and no step-by-step intrusion procedure.

## Expected input

The user provides some or all of:
- client objectives, even expressed in business terms ("protect payroll", "do not stop production");
- context: sector, size, scope, exclusions, known detection capabilities (SOC, EDR, monitoring), production constraints;
- duration, team resources, existing rules of engagement.

If the authorised scope, the exclusions or the emergency contacts are not provided, ask for them: never presume an authorisation.

## Scenario structure

For each scenario, produce in this order:

1. **Business objective**: the question the client wants answered, in one sentence.
2. **Crown jewels**: the assets or functions whose compromise would be most serious, inferred from the objectives. Mark each "Inferred" or "Confirmed by the client", and list the questions to confirm.
3. **Threat actor profile**: motivation, level of resources, plausible entry point, operating duration. Present it as a working hypothesis; attribute no real activity to a named group.
4. **ATT&CK tactics to cover**: at tactic level only, with the identifier (TA0043 Reconnaissance, TA0042 Resource Development, TA0001 Initial Access, TA0002 Execution, TA0003 Persistence, TA0004 Privilege Escalation, TA0006 Credential Access, TA0007 Discovery, TA0008 Lateral Movement, TA0009 Collection, TA0011 Command and Control, TA0010 Exfiltration, TA0040 Impact). Tactic TA0005 Defense Evasion is deliberately not covered by this skill: if the exercise is meant to test detection of low-noise actions, refer that point to the rules of engagement, without describing any technique. For each tactic kept: what the scenario seeks to validate, not how to do it. Do not write a technique identifier you are not certain of. Tactics unrelated to the objectives are dropped with a justification.
5. **Success criteria**: measurable (reaching a crown jewel shown by harmless evidence, time to detect, time to respond), and stop criteria.
6. **Detection opportunities**: for each step of the scenario, what the blue team should be able to observe (log sources, expected alerts) and the expected response. These are improvement leads, not a ranking of the blue team.
7. **Safety constraints**: prohibited actions (denial of service, destruction, modification of production data), time windows, systems and data out of scope, handling of real data limited to minimal proof.
8. **Deconfliction**: emergency contacts, communication channel between the Red Team and the client lead, immediate stop procedure, how to tell a Red Team action from a real incident, timestamped action log.

## Rules

- No evasion or concealment technique, no payload, no exploitation chain, no credential collection by phishing, no stealth persistence. If the user asks for any, decline that point and offer to describe the objective to validate and the matching detection indicator.
- Never invent an element of the client's environment (tools, headcount, architecture). What is missing becomes a question.
- The scenarios are proposals: remind that they must be validated by the client and written into the rules of engagement before any action.
- Remind not to share client information with an online AI service without contractual authorisation.
- End with a "Questions to clarify" section and an "Assumptions" section.
- Write in the user's language, with no pleasantries or generic conclusion.

A successful Red Team exercise starts from business objectives, not a list of tools. This skill helps move from "we want to know whether an attacker could reach our payroll data" to a scoped scenario: critical assets targeted, adversary emulated, tactics to cover, success criteria, detection opportunities for the blue team and safety guardrails.

What the skill produces

  • The list of crown jewels inferred from the objectives, with the questions to ask the client to confirm them.
  • An adversary profile to emulate (motivation, level of resources, plausible entry point), without claiming to attribute a real threat.
  • The ATT&CK tactics to cover, designated at tactic level (TA00xx identifiers), with what the scenario seeks to validate for each.
  • Measurable success criteria, and for each step the detection and response opportunities the blue team should be able to use.
  • The safety constraints (prohibited actions, windows, data not to be touched) and the deconfliction plan.

The skill describes no evasion technique and no payload. For general preparation, see preparing a Red Team exercise; for the contractual framework, the Red Team rules of engagement.

Install the skill

In Claude Code, for all your projects:

mkdir -p ~/.claude/skills/red-team-scenario
curl -fsSL -o ~/.claude/skills/red-team-scenario/SKILL.md \
  https://arsenal.bearops.fr/telechargements/skills/scenario-red-team/en/SKILL.md

For a single project, put the file in .claude/skills/red-team-scenario/SKILL.md at the root of the repository. With another agent, add the file’s content to its project instructions.

Use it

Give the client’s objectives as expressed, even if vague:

The client is an industrial SME (400 employees, one production site, an
outsourced SOC during business hours). Its management wants to know whether an
external attacker could disrupt production or reach the manufacturing designs.
Propose two Red Team scenarios for a four-week exercise.

[add scope, exclusions, production constraints and emergency contacts]

Limitations

  • The scenarios are proposals to validate with the client and write into the rules of engagement, not an executable attack plan.
  • Adversary profiles are working hypotheses: the skill attributes no real activity to a named group.
  • ATT&CK identifiers change: check them on the MITRE website before quoting them in a deliverable.
  • Do not send any client information (organisation, assets, people) to an online AI service unless your contract allows it.

Sources

Related resources

  • Methodology · Infrastructure and AD pentesting · Red teaming

    Active Directory config review

    Defensive Active Directory review checklist for audits: privileged accounts, krbtgt, delegation, LAPS, legacy protocols, GPOs, ADCS and logging.

    Published

  • Cheatsheet · Infrastructure and AD pentesting · Red teaming

    Active Directory enumeration

    Authenticated AD enumeration for an internal test: domain, accounts, groups, password policy, shares, LDAP queries and BloodHound collection.

    Published

  • AI skill · Web pentesting · Infrastructure and AD pentesting · Red teaming

    Engagement scoping skill

    Claude Code skill that prepares the scoping of a pentest or Red Team: client questionnaire, missing authorisations, production risks and test windows.

    Published

BEAROPS

Ready for a red team exercise?

BEAROPS runs red team exercises lasting one to six months, built on MITRE ATT&CK, to put your detection and response to the test.